2 PhDs in Trinity College Dublin, Ireland, to start in Sept 2019. Both come with a stipend of 18000Euros per year, along with full student fees for a 4 year period. Please contact me  at nharte@tcd.ie if interested in either post.

Human Speech?  How do I know it’s Real?

20 years ago, the major focus in developing speech synthesis systems was testing the intelligibility of the output speech. More recently, attention has switched focus to assessing not only intelligibility, but also naturalness, pleasantness, pauses, stress, intonation, emotion and listening effort. The intelligibility of systems is now so high, that synthetic voices are becoming more human-like. This is good news for generating realistic synthetic speech for applications such as voice reconstruction or gaming. In tandem, research in the area of speaker verification, or voice based biometrics, has started to pay closer attention to the issue of spoofing – where systems are attacked with reconstructed speech. Now, with improvements in speech synthesis, another realistic form of spoofing is the use synthetic speech generated by modelling the target user. So how can you tell when speech is real, or when it is fake? This is the focus of this PhD project and it goes to the very core of the nature of human speech.

Remote and Automatic Monitoring of Bird Populations

The objective of this PhD project is to define the next-generation approaches to the use of remote monitoring for populations of birds of conservation concern. This PhD programme will develop acoustic techniques for the monitoring of bird species of conservation concern by leveraging recent developments in speech and language processing technologies. The PhD will develop appropriate approaches to acoustic data collection in the wild to ensure that acoustic surveys yield accurate bird population data and investigate audio signal analysis steps necessary to extract useful information from these long recordings. Approaches will involve the use of signal processing and deep learning. The research will be conducted in collaboration with the Dept of Zoology at TCD.

--

Associate Professor Naomi Harte

School of Engineering,

Trinity College Dublin

www.sigmedia.tv

Back

Top

Post-doc/PhD position Pattern mining for Neural Networks debugging: application to speech recognition

Back

Top

The Department of Speech-Language Pathology, in the Faculty of Medicine at the University of Toronto, invites applications for a full-time tenure-stream appointment in the field of pediatric language disorders. The appointment will be at the rank of Assistant Professor and will commence on January 1, 2020 or shortly thereafter.

The successful candidate must have a Ph.D. in Speech-Language Pathology, Communication Disorders or a related field of research and expertise in child language and child language disorders, with a focus on semantics, syntax, pragmatics or literacy with a demonstrated record of excellence in research and teaching. Completion of a post-doctoral fellowship will be considered an asset, as will experience supervising student research and/or post-doctoral fellows. We require that candidates have a clinical background in speech-language pathology/communication disorders and are eligible for registration with the College of Audiologists and Speech-Language Pathologists of Ontario (CASLPO). We seek candidates whose research and teaching interests complement and strengthen our existing departmental strengths. The successful candidate will be expected to pursue innovative and independent research at the highest international level and to establish an outstanding, competitive and externally funded research program.

Research excellence should be demonstrated by high-quality peer-reviewed publications, peer-reviewed funding, the submitted research statement, presentations at significant conferences, awards and accolades, and strong endorsements from referees of high standing.

Evidence of excellence in teaching will be provided through teaching accomplishments and awards, the teaching dossier, including a detailed teaching statement, sample course syllabi, and teaching evaluations submitted as part of the application, as well as strong letters of reference. 

Salary will be commensurate with qualifications and experience. 

This is an exceptional opportunity to join the Department of Speech-Language Pathology at the University of Toronto, one of the most highly ranked research universities in North America. The department is housed in the Rehabilitation Sciences Building, which provides excellent teaching and research facilities. The University of Toronto offers unique opportunities for collaborative and interdisciplinary research, encourages innovative scholarship, and provides the prospect of teaching a diverse student population. 

All qualified candidates are invited to apply by clicking the link below. Applicants must submit a cover letter, curriculum vitae, research statement (up to 3 pages), copies of up to three representative publications and a teaching dossier to include a detailed teaching statement (up to 3 pages), sample syllabi, and teaching evaluations.

Applicants must also arrange to have three letters of reference sent directly by the referee via email (on letterhead and signed) to search.slp@utoronto.ca by the closing date. If you have questions about this position, please contact search.slp@utoronto.ca. All application materials must be submitted online.


Submission guidelines can be found at http://uoft.me/how-to-apply. We recommend combining attached documents into one or two files in PDF format. 

All application materials, including reference letters, must be received by July 18, 2019.

For more information about the Department of Speech-Language Pathology, please visit our home page at https://slp.utoronto.ca/. 

The University of Toronto is strongly committed to diversity within its community and especially welcomes applications from racialized persons / persons of colour, women, Indigenous / Aboriginal People of North America, persons with disabilities, LGBTQ persons and others who may contribute to the further diversification of ideas. 

As part of your application, you will be asked to complete a brief Diversity Survey. This survey is voluntary. Any information directly related to you is confidential and cannot be accessed by search committees or human resources staff. Results will be aggregated for institutional planning purposes. For more information, please see http://uoft.me/UP.

All qualified candidates are encouraged to apply; however, Canadians and permanent residents will be given priority.

Back

Top

SUBJECT TITLE: Bio-Bayes Predictions -- Coupling Biological and Bayesian Predictive
Models in Neurocognitive Speech Processing
RESEARCH FIELD: Computer science, Cognitive Science, Psychological sciences, Neurosciences
SCIENTIFIC DEPARTMENT (LABORATORY?S NAME): LPNC -- Laboratoire de Psychologie et
NeuroCognition
DOCTORAL SCHOOL?S: MSTII -- Mathématiques, Sciences et technologies de l'information,
Informatique SUPERVISOR?S NAME: Julien Diard, PhD

SUBJECT DESCRIPTION:
The issue of the relationship between perception and production mechanisms is central to
many domains in cognitive science. In speech communication for instance, where
predictions from speech production simulation interact in various ways with perceptual
processes. In this context, we have developed COSMO (Communicating about Objects using
SensoriMotor Operations), a family of Bayesian algorithmic models of communicating
agents. We have previously used such models to study the evolution of phonological
systems (Moulin-Frier et al., 2015), speech perception and learning (Laurent et al.,
2017; Barnaud et al., 2017; 2018), and speech production and adaptation (Patri et al.,
2015; 2018).

However, so far, these models consider greatly simplified temporal dimensions. For
instance, syllable perception was restricted to consonant-vowel syllables, assuming that
the key points of speech trajectories, respectively at the center of the consonant and
the vowel, were previously identified. This, of course, contrasts with natural speech
processing, where sensory inputs and motor controls continuously unfold over time.
Indeed, the neuronal substrates, in the brain, that deal with auditory input are well
described in terms of their oscillatory nature, since they intrinsically have to deal
with temporal properties of speech, and their predictive nature, since they aim at
anticipating events.

In this PhD project, we aim at extending the COSMO framework to define the first Bayesian
perceptuo-motor model of continuous speech communication. In previous work, in the domain
of Bayesian visual word recognition modeling (Phénix et al., 2018; Ginestet et al.,
2019), we have developed mathematical tools to describe the temporal dynamics of
perceptual evidence accumulation across layers of hierarchical representations.
Probability distributions at each layer (letters and words) evolve continuously over
time, as a function of bottom-up sensory evidence and top-down lexical constraints, to
predict upcoming events. Crucially, we have developed mathematical tools to model, on the
one hand, attentional control of these information flows, and, on the other hand,
asynchronous and asymmetric information transfer. Applying these mathematical constructs
to speech communication modeling would yield a novel class of Bayesian hierarchical and
predictive models, able to account for observations of neuronal oscillatory systems in
the brain.

This PhD project is part of an international collaboration with Anne-Lise Giraud and
Itsaso Olasagasti of the ?Auditory, Speech and Language Neuroscience? group at UNIGE
(Université de Genève, Switzerland), with regular meetings and visits to Geneva planned.
This international collaboration with Geneva will provide a unique framework for mixing
the Bayesian approach with neuroscience constraints and data, providing a valuable
multidisciplinary environment for the PhD project. The PhD candidate will integrate the
?Language? research team at LPNC (Laboratoire de Psychologie et NeuroCognition,
Grenoble), will be supervised by Julien Diard (CR CNRS, LPNC) and Jean-Luc Schwartz (DR
CNRS, Gipsa- Lab), and will be registered in the MSTII (Mathématiques, Sciences et
technologies de l'information, Informatique) doctoral school of Univ. Grenoble Alpes.

ELIGIBILITY CRITERIA
Applicants must hold a Master's degree (or be about to earn one) or have a university
degree equivalent to a European Master's (5-year duration),
Applicants will have to send an application letter in English and attach: - their last
diploma;
- their CV;
- letters of recommendation are welcome.
Address to send their application: Julien.Diard@univ-grenoble-alpes.fr

SELECTION PROCESS
Application deadline: June, 21, 2019 at 17:00 (CET) Applications will be evaluated
through a three-step process:
1. Eligibility check of applications before June, 24, 2019
2. 1st round of selection: the applications will be evaluated by a Review Board between
June, 24, and June, 28,
2019. Results will be given June, 28, 2019.
3. 2nd round of selection: shortlisted candidates will be invited for an interview
session in Grenoble on July, 5,
2019 (in Grenoble or by Skype).

TYPE of CONTRACT: temporary-3 years of doctoral contract JOB STATUS: Full time
HOURS PER WEEK: 35
OFFER STARTING DATE: October, 1st, 2019
APPLICATION DEADLINE: June, 21, 2019
Salary: between 1768.55 ? and 2100 ? brut per month (depending on complementary activity
or not)
PROFILE SUMMARY:
The candidate must have a background in computer science, applied mathematics or signal
processing, with a strong affinity for cognitive sciences, or instead a background in
cognitive science or computational neuroscience, with a strong affinity for mathematical
modelling. This training must be validated, or be validated soon (before Summer 2019), by
a Masters level diploma (or equivalent).

The candidate must have previous experience in scientific research in a laboratory.
Mastery of a scientific calculus language, or of a general-purpose programming language
is required (R, Python, Matlab, etc.). The candidate must also have a good command of
scientific English, both spoken and written.

Previous experience in probabilistic modeling, dynamic systems modeling, or connectionist
modeling are a plus but are not required. Previous knowledge of the field of perception
modeling or speech production is also a plus.

 

Back

Top

Lieu de travail : Grenoble

Date de publication : 1 Mai 2019

Noms des responsables scientifiques : Gérard Bailly (DR CNRS, GIPSA-Lab, U. Grenoble Alpes) et Pascal Huguet (DR CNRS, LAPSCO, U. Clermont Auvergne)

Type de contrat : CDD Doctorant/Contrat doctoral

Durée du contrat : 36 mois

Date de début de la thèse : 1 octobre 2019

Quotité de travail : Temps complet

Rémunération : 2 135,00 € brut mensuel

Sujet de la these:

Impact de la présence de robots attentifs et bienveillants sur les comportements et fonctionnements cognitifs humains

Contexte

La robotique sociale humanoïde vise à créer des robots capables d’interagir avec des partenaires humains à des fins de coopération dans des secteurs aussi variés que l’assistance aux personnes âgées, l’apprentissage de compétences par les enfants ou encore la cobotique dans l’industrie 4.0. L’objectif est souvent de substituer à un partenaire humain ou animal le robot dans des tâches où les propriétés d’endurance, de rapidité ou de flexibilité des attributions sociales de ces avatars sociaux sont bénéfiques. Si beaucoup d’études en interaction face-à-face montrent l’impact des comportements verbaux [1] et coverbaux [2] des robots sociaux sur ceux de leurs partenaires humains, peu d’études ont été consacrées aux effets comparés entre présence robotique vs. humaine en matière de surveillance de tâches dans lesquelles le robot ou son modèle humain ne sont pas directement engagés. Deux études récentes conduites au LAPSCO [3] [4] ont notamment permis de répliquer avec succès, sous condition de présence robotique, l’influence généralement bénéfique de la présence humaine en matière de contrôle de l’attention dans des tâches impliquant de réprimer un automatisme cognitif néfaste dans l’activité cible (e.g. [5]).

Sujet et plan de travail

Nous proposons ici d’explorer l’impact du comportement d’un robot humanoïde possédant des capacités de communication verbale et co-verbale étendues [6] sur les performances (en matière de contrôle cognitif) de sujets impliqués dans des tâches dans lesquelles le robot est directement impliqué ou plus simplement présent dans l’environnement d’interaction. Nous allons donc varier le degré avec lequel le robot s’immisce dans la tâche principale, afin de déterminer les signaux susceptibles d’optimiser l’expression des bénéfices attachés aux robots sociaux humanoïdes (et à terme, leur acceptabilité) et les signaux plus néfastes à éliminer.

Dans le cadre de cette thèse, nous explorerons l’impact du comportement robotique – dans un premier temps, téléopéré par un pilote humain [7] puis piloté par les modèles appris sur ces données comportementales [8] – sur le comportement observable des sujets (e.g., performances cognitives, signaux verbaux et co-verbaux, notamment regard), sur leurs signaux physiologiques (i.e. conduction dermale, rythme respiratoire et cardiaque) et sur leurs activités cérébrales (i.e. étude de l’onde négative d’erreur par EEG).

Retombées scientifiques et technologiques

Les retombées attendues de cette thèse sont multiples : d’abord des modèles de prédiction des liens entre comportements observables, variables physiologiques et activités cérébrales sous-jacentes, ceci en fonction des performances observées et des comportements actifs et réactifs du robot ; ensuite, un ensemble de stratégies de comportement de robots attentifs et bienveillants, adaptées aux profils psychologiques des sujets et aux indicateurs de performance propres à la tâche ; et enfin des protocoles d’évaluation automatique de ces profils psychologiques par des robots sociaux interactifs, susceptibles d’être déployés dans des applications en santé, éducation ou évaluation de ressources humaines.

Références

[1] K.-Y. Chin, Z.-W. Hong, and Y.-L. Chen, “Impact of using an educational robot-based learning system on students’ motivation in elementary education,” IEEE Transactions on learning technologies, vol. 7, no. 4, pp. 333–345, 2014.

[2] S. Andrist, X. Z. Tan, M. Gleicher, and B. Mutlu, “Conversational gaze aversion for humanlike robots,” presented at the Proceedings of the 2014 ACM/IEEE international conference on Human-robot interaction, 2014, pp. 25–32.

[3] N. Spatola et al., “Improved Cognitive Control in Presence of Anthropomorphized Robots,” International Journal of Social Robotics, pp. 1–14, 2019.

[4] N. Spatola et al., “Not as bad as it seems: When the presence of a threatening humanoid robot improves human performance,” Science Robotics, vol. 3, no. 21, eaat5843, 2018.

[5] D. Sharma, R. Booth, R. Brown, and P. Huguet, “Exploring the temporal dynamics of social facilitation in the Stroop task,” Psychonomic bulletin & review, vol. 17, no. 1, pp. 52–58, 2010.

[6] Parmiggiani, Alberto, Elisei, Frédéric, Maggiali, Marco, Randazzo, Marco, Bailly, Gérard, and Metta, Giorgio, “Design and validation of a talking face for the iCub,” International Journal of Humanoid Robotics, vol. 12, no. 3, 20 pages, 2015.

[7] R. Cambuzat, F. Elisei, G. Bailly, O. Simonin, and A. Spalanzani, “Immersive teleoperation of the eye gaze of social robots,” in Int. Symposium on Robotics (ISR), Munich, Germany, 2018, pp. 232–239.

[8] Nguyen, Duc-Canh, Bailly, Gérard, and Elisei, Frédéric, “Learning off-line vs. on-line models of interactive multimodal behaviors with Recurrent Neural Networks,” Pattern Recognition Letters (PRL), pp. 29–36, 2017.

Au sein du département Parole & Cognition du GIPSA-Lab (Grenoble, France), l'équipe Cognitive Robotics, Interactive Systems & Speech Processing (CRISSP) développe des modèles de comportements multimodaux (parole, gestes, regard, etc.) pour des robots humanoïdes en interaction avec des partenaires humains. Elle s'appuie sur les moyens techniques du GIPSA-Lab et en particulier sur la plate-forme robotique MICAL qui gère le robot iCub NINA et les outils de développement.

Cette thèse s'inscrit dans le projet de l'équipe de développer des robots engagés dans des tâches finalisées nécessitant un contrôle fin de l’engagement. Elle est financée par le projet 80 ans du CNRS «Robotique Sociale Humanoïde et Cognition » (RSHC) impliquant des chercheurs de trois laboratoires : le LAPSCO et le LIMOS à Clermont et le GIPSA-Lab à Grenoble,

La thèse sera rattachée à l’école doctorale EEATS de Grenoble. Des déplacements et trois séjours de courte-durée à Clermont-Ferrand seront à prévoir pour collaborer avec les chercheurs du LAPSCO et du LIMOS de l’UCA.

Le candidat devra être titulaire d’un diplôme d’ingénieur et/ou d’un master en Sciences Cognitives ou en Robotique d’interaction. Le poste nécessite de solides connaissances en expérimentation, programmation et analyse statistique. Une formation initiale en neurosciences est un plus. Le candidat doit avoir de bonnes aptitudes de communication orale et écrite (français et anglais nécessaires) pour présenter aux congrès et rédiger des articles dans des revues scientifiques. Nous recherchons un jeune chercheur qui saura s’impliquer dans son projet, curieux, ayant une certaine autonomie et une forte motivation pour développer des compétences en synthèse et évaluation de comportements dans le domaine de l’interaction homme-robot. De plus, le candidat devra être apte à travailler en équipe sur des projets pluridisciplinaires.

Les candidatures devront inclure un CV détaillé ; au moins deux références (personnes susceptibles d’être contactées) ; une lettre de motivation d’une page ; un résumé d’une page du mémoire de master ; les notes de Master 1 ou 2 ou d’école d’ingénieur.

La date limite pour l’envoi des candidatures est le 15/7/2019

Back

Top

Researcher position (W/M) at IRCAM

Fixed-term contract of 12 months

INTRODUCTION TO IRCAM:

IRCAM is a leading non-profit organization associated to Centre Pompidou, dedicated to music production,

scientific research and education in sound and music technologies. In the joint research unit, UMR 9912

STMS (Science and Technology for Music and Sound), supported by IRCAM, Sorbonne University, and the

CNRS, specialized research teams conduct research and development in the areas of acoustics, audio

signal processing, human-computer interaction, computer music and musicology. IRCAM is located in the

centre of Paris near the Centre Pompidou, at 1, Place Igor Stravinsky 75004 Paris.

POSITION DESCRIPTION:

IRCAM is looking for a researcher for the development of music content analysis technologies (such as

instrument recognition, audio quality, auto-tagging) in the framework of a technology transfer to the

Universal Music Group.

The hired researcher will also collaborate with the development team and participate in the project activities

(evaluation of technologies, meetings, specifications, reports).

REQUIRED EXPERIENCES ET SKILLS:

 Very high skills in audio signal processing (spectral analysis, audio-feature extraction, parameter

estimation) (the candidate should preferably hold a PhD in this field) ;

 Very high skills in machine learning (SVM, ConvNet) (the candidate should preferably hold a PhD in

this field) ;

 Good skills in distributed computing ;

 High-skills in Matlab and Python programming, skills in C/C++ programming ;

 Good knowledge of UNIX environments (GNU-Linux ou MacOSX) ;

 High productivity, methodical works, excellent programming style.

SALARY: According to background and experience.

Deadline for application: June, 17th, 2019

Please send an application letter with the reference 201906UMGRES together with your resume and any

suitable information addressing the above issues preferably by email to: mignot at ircam dot fr with cc to

vinet at ircam dot fr and roebel at ircam dot fr.

Back

Top

PhD title: Decentralized collaborative learning for speech recognition in

a context of privacy protection

PhD project

The candidate will contribute to the DEEP-PRIVACY project funded by the French national

research agency (ANR). DEEP-PRIVACY aims at developing private-by-design acoustic

models for use in automatic speech recognition systems. Indeed, more and more devices

–mobile or not– invest our daily lives. These devices use voice technology to interact with

human users who can access IT services through natural interaction (Apple Siri, Amazon

Alexa, Google Home ...). With the stated goal of improving the quality of their services,

which includes automatic speech recognition, companies that offer such solutions usually

collect private audio data from these devices. This constitutes a real risk for privacy.

The PhD proposal is to study and propose approaches for distributed and decentralized

learning of neural acoustic models in which user data remain on the user's device.

A first objective consists in experimenting and comparing different approaches from the

literature to the particular case of acoustic models in speech processing tasks. We will

focus on machine learning algorithms where data are collected locally on every peer and

are not transmitted to a central server. Communications are restricted to models or

updates of weights computed locally on the device. The cases of centralized federated

learning approaches [Konečny et al. 2016, Leroy et al. 2018] and purely decentralized

approaches [Bellet et al. 2017] will be studied. Comparisons will include several trade-offs

involving model size and communication costs for instance.

Beyond the implementation and experimentation of federated learning and distributed

collaborative learning approaches for automatic speech recognition, a study on the nature

of the information conveyed (paralinguistics, phonetics, lexics) during these exchanges will

also be conducted to assess the level of privacy protection provided by the proposed

approaches. More precisely, it will be a question of studying the possibility of recognizing,

at a level that will have to be quantified if necessary, the speaker and the phonemes, even

the words uttered, by analyzing the values of the exchanged updates.

We will also investigate how to locally adapt the acoustic models to the voice and speech

characteristics of the user in order to obtain personalized models. The amount of data and

adaptation time will be studied.

In general, the thesis will keep a critical eye on the advantages and disadvantages offered

by collaborative automatic learning in a real application framework [Bhowmick et al. 2018;

Hitaj et al. 2018]. In view of the results of these experiments and analyses, the purpose of

the thesis will be to propose effective solutions to overcome the disadvantages and

propose an effective framework, linking privacy protection and improvement of acoustic

modeling for speech recognition in a distributed deployment context.

Technical skills required:

•Master’s degree in machine learning or in computer science

•Background in deep learning, and in statistics

•Experience with deep learning tools is a plus (preferably PyTorch)

•Good programing skills (preferably in Python)

•Experience in speech and/or speaker recognition is a plus

General information

This PhD thesis fits within the scope of a collaborative project (project DEEP-PRIVACY,

funded by the French National Research Agency) involving the LIA (Laboratoire

Informatique d’Avignon), the MAGNET team of Inria Lille - Nord Europe, the LIUM

(Laboratoire d'Informatique de l'Université du Mans) and the MULTISPEECH team of Inria

Nancy - Grand-Est.

This PhD position is in collaboration with Avignon University, and will be co-supervised by

Yannick Estève (https://cv.archives-ouvertes.fr/yannick-esteve) and Marc Tommasi (http://

researchers.lille.inria.fr/tommasi). The work location will mainly be in Avignon (LIA, Avignon

University), with some stays in Lille (CRIStAL / INRIA Lille Nord Europe).

Knowledge of the French language is not required.

Additional information:

Three-year work contract, with a monthly net salary of approximately 1685€/month, and a

financial support for international research training and conference participation plus a

contribution to the research costs.

Duration: 3 years

Starting date: September or October 2019

Contacts:

Yannick ESTÈVE yannick.esteve@univ-avignon.fr

Marc TOMMASI marc.tommasi@inria.fr

Doctoral School: Sciences et Agrosciences (Avignon University)

References

[Bellet et al. 2017] Bellet, A., Guerraoui, R., Taziki, M., & Tommasi, M. (2017). Personalized and private peerto-

peer machine learning. arXiv preprint arXiv:1705.08435

[Bhowmick et al. 2018] A. Bhowmick, J. Duchi, J. Freudiger, G. Kapoor, and R. Rogers. Protection Against

Reconstruction and Its Applications in Private Federated Learning. arXiv preprint https://arxiv.org/abs/

1812.00984

[Hitaj et al. 2018] B. Hitaj, G. Ateniese, F. Perez-Cruz. Deep Models Under the GAN: Information Leakage

from Collaborative Deep Learning. arXiv preprint https://arxiv.org/abs/1702.07464

[Konečný et al. 2016] J. Konečný, H. B. McMahan, F. X. Yu, P. Richtárik, A. T. Suresh, & D. Bacon. (2016).

Federated learning: Strategies for improving communication efficiency. arXiv preprint arXiv: https://arxiv.org/

abs/1610.05492

[Leroy et al. 2018] D. Leroy, A. Coucke, T. Lavril, T. Gisselbrecht, J. Dureau, Federated Learning for Keyword

Spotting. arXiv preprint https://arxiv.org/abs/1810.05512

Back

Top

Modélisation multimodale de l’expressivité et de l’alignement pour l’interaction humain-machine

Directrice de thèse : Catherine Pelachaud (ISIR)

Co-encadrant : Nicolas Obin (STMS)

Contexte

Cette thèse s’inscrit dans un contexte particulièrement riche en développement d’interfaces de communication entre l’humain et la machine. Par exemple, l’émergence et la démocratisation des assistants personnels (smartphones, home assistants, chatbots) font de l’interaction avec la machine une réalité quotidienne pour de plus en plus d’individus. Cette pratique tend à s’amplifier et à se généraliser à un grand nombre d’usages et de pratique de l’être humain : depuis les agents d’accueil (aujourd’hui, quelques robots Pepper plus pour la démo que pour un usage réel), la consultation à distance, ou les agents embarqués dans les véhicules autonomes. L’expansion des usages appelle à une extension des modalités d’interaction et à l’amélioration de la qualité de l’interaction avec la machine : aujourd’hui, la voix constitue la modalité privilégiée de l’interaction, et les scénarios d’interaction demeurent très limités (demande d’information, question-réponse, pas de réelle interaction dans la durée). Les limitations principales sont d’une part une faible expressivité : le comportement de l’agent est encore souvent monomodal (voix, comme les assistants Alexa ou Google Home) et demeure très monotone, ce qui limite grandement l’acceptabilité, la durée et la qualité de l’interaction ; et d’autre part le comportement de l’agent est peu ou pas adapté à l’interlocuteur, ce qui diminue l’engagement de l’humain dans l’interaction. Lors d’une interaction humain-humain les phénomènes d’alignement (e.g., ton de voix, vitesse de mouvement corporel) sont des indices de compréhension commune et d’engagement dans l’interaction (Pickering et Garrod, 1999; Castellano et al, 2012). L’engagement est marqué par des comportements non-verbaux sociaux (nonverbal social behaviors) à des moments spécifiques de l’interaction : ce peut être des signaux de feedbacks (pour indiquer être en phase avec l’interactant), ou bien une forme d’imitation (par exemple : un sourire appelle un autre sourire, le ton de la voix reprend des éléments de celui de l’interactant), ou encore des signaux synchronisés avec ceux de l’interactant (la gestion des tours de parole). Cette thèse vise à modéliser le comportement de l’agent en fonction de celui de l’utilisateur pour qu’il puisse montrer son engagement attentionnel en vue de maintenir l’interaction et de rendre ses messages plus compréhensifs. L’adaptation du comportement de l’agent se produira à différents niveaux comportementaux (prosodique, lexicale, comportementale, imitation, tour de parole…). L’interaction humain-machine, avec un fort potentiel applicatif dans de nombreux domaines, est un exemple d'interdisciplinarité nécessaire entre humanités numériques, robotique, et intelligence artificielle.

Objectif

L’objectif de la thèse est de mieux comprendre et à modéliser les mécanismes qui régissent l’interaction multimodale (voix et geste) entre un humain et une machine, pour permettre de lever des verrous technologiques et permettre d’élaborer un agent conversationnel capable de s’adapter de manière naturelle et cohérente à un interactant humain.

1) Expressifs (Léon, 1993) : capable d’avoir une expression variée et cohérente pour maintenir l’attention de l’interlocuteur, souligner les points importants, améliorer la qualité de l’interaction et en allonger la durée (dépasser un ou deux tours de parole)

2) Alignés sur le comportement multimodal de l’interlocuteur (Pickering et Garrod, 1999; Castellano et al, 2012; Clavel et al, 2016) : c’est-à-dire capable d’adapter son comportement en fonction du comportement de l’interlocuteur, pour renforcer l’engagement de ce dernier dans l’interaction.

Dans un premier temps, la thèse proposera de réaliser une architecture neuronale unifiée pour la modélisation générative du comportement multimodale de l’agent. L’expressivité de l’agent, prosodique (Obin, 2011; Obin, 2015) et gestuelle (Pelachaud, 2009), sera modélisée par des architectures neuronales récurrentes aujourd'hui couramment utilisées pour la voix et le geste (Bahdanau et al, 2014, Wang, 2017, Robinson & Obin, 2019). La thèse se focalisera sur deux aspects essentiels de la modélisation du comportement de l’agent : le développement d’architectures structurées sur plusieurs échelles temporelles pour améliorer la modélisation de la variabilité prosodique et gestuelle à l’échelle de la phrase et à l’échelle du discours (Le Moine & Obin, 2019), et l’apprentissage d’un comportement multimodal cohérent par l’approfondissement de mécanismes d’attention multimodaux partagés appliqués à la synchronicité des profils prosodiques et gestuels générés (He, 2018).

Dans un deuxième temps, la thèse s’attaquera à l’alignement du comportement de l’agent avec celui de l’humain. La thèse approfondira particulièrement l’apprentissage interactif et par imitation pour adapter de manière cohérente le comportement multimodal de l’agent à l’humain (Weber, 2018; Mancini, 2019), à partir des bases de données de dialogues accessibles (telles que NoXi (récoltées à l’ISIR et annotées en terme d’engagement), IEMOCAP (USC, Carlos Busso), Gest-IS (Edinburgh University, Katya Saint-Amard)) pour apprendre la relation et aussi leur adaptation au cours de l'interaction entre les profils prosodiques et comportementaux des interlocuteurs.

La thèse sera co-encadrée par Catherine Pelachaud, de l’équipe PIRoS de l’ISIR, spécialisée en interaction humain-machine et agents conversationnels, et par Nicolas Obin, de l’équipe Analyse et Synthèse des Sons (AS) de STMS, spécialisée en modélisation générative de signaux de parole.. Le doctorant bénéficiera par ailleurs des connaissances, savoir-faire, et outils existants à STMS et à l’ISIR (par exemple : synthétiseur de parole ircamTTS développé à STMS, plateforme GRETA développée à l’ISIR) et de la logistique de calcul de STMS (serveur de calculs, GPU).

Bibliographie

(Bevacqua et al., 2012) Elisabetta Bevacqua, Etienne de Sevin, Sylwia Julia Hyniewska, Catherine Pelachaud, A listener model: Introducing personality traits, Journal on Multimodal User Interfaces, special issue Interacting ECAs, Elisabeth André, Marc Cavazza and Catherine Pelachaud (Guest Editors), July 2012, 6(1-2), pp 27-38.

(Castellano et al., 2012) G. Castellano, M. Mancini, C. Peters, P. W. McOwan. Expressive copying behavior for social agents: a perceptual analysis. IEEE Trans Syst, Man Cybern, Part A: Syst Hum 42(3), 2012.

(Clavel et al., 2016) Chloé Clavel, Angelo Cafaro, Sabrina Campano, and Catherine Pelachaud, Fostering user engagement in face-to-face human-agent interactions, in A. Esposito and L. Jain (Eds), Toward Robotic Socially Believable Behaving Systems - Volume I: Modeling Social Signals, Springer Series on Intelligent Systems Reference Library (ISRL), 2016

(Glas and Pelachaud, 2015) N. Glas, C. Pelachaud, Definitions of Engagement in Human-Agent Interaction, workshop ENHANCE, in International Conference on Affective Computing and Intelligent Interaction (ACII), 2015.

(Hall et al., 2005) L. Hall, S. Woods, R. Aylett, L. Newall, A. Paiva. Achieving empathic engagement through affective interaction with synthetic characters. Affective computing and intelligent interaction, 2005.

(He, 2018) Xiaodong He, Deep Attention Mechanism for Multimodal Intelligence: Perception, Reasoning, & Expression across Language & Vision, Microsoft Research, AI NEXTCon, 2018.

(Le Moine & Obin, 2019) Clément Lemoine, Modélisation neuronale de l’expressivité pour la transformation de la voix, stage de Master, 2019.

(Léon, 1993) P. Léon. Précis de phonostylistique : Parole et expressivité. Paris:Nathan, 1993.

(Obin, 2011) N. Obin. MeLos: Analysis and Modelling of Speech Prosody and Speaking Style, PhD. Thesis, Ircam-Upmc, 2011.

(Obin, 2015) N. Obin, C. Veaux, P. Lanchantin. Exploiting Alternatives for Text-To-Speech Synthesis: From Machine to Human, in Speech Prosody in Speech Synthesis: Modeling and generation of prosody for high quality and flexible speech synthesis. Chapter 3: Control of Prosody in Speech Synthesis, p.189-202, Springer Verlag, February, 2015.

(Ochs et al., 2008) M. Ochs, C. Pelachaud, D. Sadek, An Empathic Virtual Dialog Agent to Improve Human-Machine Interaction, Seventh International Joint Conference on Autonomous Agents and Multi-Agent Systems (AAMAS), Estoril Portugal, May 2008.

(Paiva et al., 2017) A. Paiva, I. Leite, H. Boukricha, Hana I. Wachsmuth 'Empathy in Virtual Agents and Robots: A Survey.', ACM Trans. Interact. Intell. Syst. (2017), 7 (3):11:1-11:40.

(Pelachaud, 2009) C. Pelachaud, Studies on Gesture Expressivity for a Virtual Agent, Speech Communication, special issue in honor of Björn Granstrom and Rolf Carlson, 51 (2009) 630-639.

(Poggi, 2007) I. Poggi. Mind, hands, face and body: a goal and belief view of multimodal communication. Weidler, Berlin, 2007.

(Robinson & Obin, 2019) C. Robinson, N. Obin, A. Roebel. Sequence-to-sequence modelling of F0 for speech emotion conversion, in IEEE International Conference on Audio, Signal, and Speech Processing (ICASSP), 2019.

(Sadoughi et al., 2017) Najmeh Sadoughi, Yang Liu, and Carlos Busso, 'Meaningful head movements driven by emotional synthetic speech,' Speech Communication, vol. 95, pp. 87-99, December 2017.

(Sidner and Dzikovska, 2002) C. L. Sidner, M. Dzikovska. Human-robot interaction: engagement between humans and robots for hosting activities. In: IEEE int conf on multimodal interfaces, 2002.

(Wang, 2017) Xin Wang, Shinji Takaki, Junichi Yamagishi. An RNN-Based Quantized F0 Model with Multi-Tier Feedback Links for Text-to-Speech Synthesis, Interspeech, 2017

(Wang, 2018) Yuxuan Wang, Daisy Stanton, Yu Zhang, RJ Skerry-Ryan, Eric Battenberg, Joel Shor, Ying Xiao, Fei Re, Ye Jia, Rif A. Saurous. « Style Tokens: Unsupervised Style Modeling, Control and Transfer in End-to-End Speech Synthesis », 2018.

(Weber, 2018) K. Weber, H. Ritschel, I. Aslan, F. Lingenfelser, E. André, How to Shape the Humor of a Robot - Social Behavior Adaptation Based on Reinforcement Learning, ACM International Conference on Multimodal Interaction, 2018.

(Mancini, 2019) M. Mancini, B. Biancardi, S. Dermouche, P. Lerner, C. Pelachaud, Managing Agent’s Impression Based on User’s Engagement Detection, Proceedings of the 19th ACM International Conference on Intelligent Virtual Agents, 2019.

Back

Top

Thesis title

Privacy preserving and personalized transformations for speech recognition

This PhD thesis fits within the scope of a collaborative project (funded by the French National Research Agency) involving several French teams, among which, the MULTISPEECH team of Inria Nancy - Grand-Est and the LIUM (Laboratoire d'Informatique de l'Université du Mans).

This PhD position is in collaboration between the Multispeech team of the LORIA laboratory (Nancy) and Le Mans University. The thesis will be co-supervised by Denis Jouvet (https://members.loria.fr/DJouvet/) and Anthony Larcher (https://lium.univlemans.fr/team/anthony-larcher/). The selected candidate is expected to spend time in both teams over the course of the PhD.

Scientific Context

Over the last decade, great progress has been made in automatic speech recognition [Saon et al., 2017; Xiong et al., 2017]. This is due to the maturity of machine learning techniques (e.g., advanced forms of deep learning), to the availability of very large datasets, and to the increase in computational power. Consequently, the use of speech recognition is now spreading in many applications, such as virtual assistants (as for instance Apple’s Siri, Google Now, Microsoft’s Cortana, or Amazon’s Alexa) which collect, process and store personal speech data in centralized servers, raising serious concerns regarding the privacy of the data of their users. Embedded speech recognition frameworks have recently been introduced to address privacy issues during the recognition phase: in this case, a (pretrained) speech recognition model is shipped to the user's device so that the processing can be done locally without the user sharing its data. However, speech recognition technology still has limited performance in adverse conditions (e.g., noisy environments, reverberated speech, strong accents, etc.) and thus, there is a need for performance improvement. This can only be achieved by using large speech corpora that are representative of the actual users and of the various usage conditions. There is therefore a strong need to share speech data for improved training that is beneficial to all users, while preserving the privacy of the users, which means at least keeping the speaker identity and voice characteristics private1.

1 Note that when sharing data, users may want not to share data conveying private information at the linguistic level (e.g., phone number, person name, …). Such privacy aspects also need to be taken into account, but they are out-of-the scope of this thesis.

Missions: (objectives, approach, etc.)

Within this context, the objective of the proposed thesis is twofold. First, it aims at finding a privacy preserving transform of the speech data, and, second, it will investigate the use of additional personalized transforms, that can be applied on the user’s terminal, to increase speech recognition performance.

In the proposed approach, the device of each user will not share its raw speech data, but a privacy preserving transformation of the user speech data. In such approach, some private computations will be handled locally, while some cross-user computations may be carried out on a server using the transformed speech data, which protect the speaker identity and some of his/her features (gender, sentiment, emotions...). More specifically, this will rely on a representation learning to separate the features of the user data that can expose private information from generic ones useful for the task of interest, i.e., here, the recognition of the linguistic content. We will build upon ideas of Generative Adversarial Networks (GANs) for proposing such a privacy preserving transform. Since a few years, GANs are getting more and more used in deep learning. They

typically rely on both a generative network and a discriminative network, where the generator aims to output samples that the discriminator cannot distinguish from the true samples [Goodfellow et al., 2014; Creswell et al., 2018]. They have also been used as autoencoders [Makhzani et al., 2015] which are made of three mains blocks: encoder, generator and discriminator. In our case, the discriminators shall focus on discriminating between speakers and/or between voice-related classes (defined according to gender, emotions, etc). The training objective will be to maximize the speech recognition performance (using the privacy preserving transformed signal) while minimizing the available speaker or voice-related information measured by the discriminator.

As devices are getting more and more personal, it creates opportunities to make speech recognition more personalized. This includes two aspects: adapting the model parameters to the speaker (and to the device) and introducing personalized transforms to help hiding the speaker voice identity. Both aspects will be investigated. Voice conversion approaches provide example of transforms aiming at modifying the voice of a speaker so that it sounds like the voice of another target speaker [e.g., Chen et al., 2014; Mohammadi & Kain, 2014]. Similar approaches can thus be applied to map speaker specific features to those of a standard (or average) speaker, which thus would help concealing its identity. To take benefit of the increased personal usage of terminals, speaker and environment specific adaptation will be investigated to improve speech recognition performance. Collaborative learning mixing speech and speaker recognition has been shown to benefit both tasks [Liu et al. 2018; Garimella et al. 2015] and provide a way to combine both information in a single framework. This approach will be compared to adaptation of deep neural networks-based models [e.g., Abdel-Hamid & Jiang, 2013] to handle best different amounts of adaptation data.

Skills and profile:

Master in machine learning or in computer science

Background in statistics, and in deep learning

Experience with deep learning tools is a plus

Good computer skills (preferably in Python)

Experience in speech and/or speaker recognition is a plus

Bibliography:

[Abdel-Hamid & Jiang, 2013] Abdel-Hamid, O., & Jiang, H. Fast speaker adaptation of hybrid NN/HMM model for speech recognition based on discriminative learning of speaker code. In ICASSP-2013, IEEE International Conference on Acoustics, Speech and Signal Processing, pp. 7942-7946, 2013.

[Chen et al., 2014] Chen, L. H., Ling, Z. H., Liu, L. J., & Dai, L. R. Voice conversion using deep neural networks with layer-wise generative training. TASLP-2014, IEEE/ACM Transactions on Audio, Speech and Language Processing, 22(12), pp. 1859-1872, 2014.

[Creswell et al., 2018] Creswell, A., White, T., Dumoulin, V., Arulkumaran, K., Sengupta, B., and Bharath, A. A. Generative adversarial networks: An overview. IEEE Signal Processing Magazine 35, 1, 53-65, 2018.

[Garimella et al. 2015] Garimella, S., Mandal, A., Strom, N., Hoffmeister, B., Matsoukas, S., & Parthasarathi, S. H. K., Robust i-vector based adaptation of DNN acoustic model for speech recognition. In INTERSPEECH, 2015.

[Goodfellow et al., 2014] Goodfellow, I., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., Courville, A., and Bengio, Y. Generative adversarial nets. In Advances in neural information processing systems, pp. 2672-2680, 2014.

[Liu et al. 2018] Y. Liu, L. He, J. Liu, and M. Johnson, Speaker Embedding Extraction with Phonetic Information,' in INTERSPEECH , pp. 2247-2251, 2018

[Makhzani, 2015] Makhzani, A., Shlens, J., Jaitly, N., Goodfellow, I., and Frey, B. Adversarial autoencoders. arXiv preprint arXiv:1511.05644, 2015.

[Mohammadi & Kain, 2014] Mohammadi, S. H., & Kain, A. Voice conversion using deep neural networks with speaker-independent pre-training. In SLT-2014, Spoken Language Technology Workshop , pp. 19-23, 2014.

[Saon et al., 2017] G. Saon, G. Kurata, T. Sercu, K. Audhkhasi, S. Thomas, D. Dimitriadis, X. Cui, B. Ramabhadran, M. Picheny, L.-L. Lim, B. Roomi, and P. Hall. English conversational telephone speech recognition by humans and machines. Technical report, arXiv:1703.02136, 2017.

[Xiong et al., 2017] W. Xiong, J. Droppo, X. Huang, F. Seide, M. Seltzer, A. Stolcke, D. Yu, and G. Zweig. Achieving human parity in conversational speech recognition. Technical report, arXiv:1610.05256, 2017.

Additional information:

Supervision and contact:

Denis Jouvet (denis.jouvet@loria.fr; https://members.loria.fr/DJouvet/)

Anthony Larcher (anthony.larcher@univ-lemans.fr; https://lium.univlemans.fr/team/anthony-larcher/)

Additional link:

Ecole Doctorale IAEM Lorraine (http://iaem.univ-lorraine.fr/)

Duration: 3 years

Starting date: autumn 2019

The candidates are required to provide the following documents in a single pdf or ZIP file:

 CV

 A cover/motivation letter describing their interest in the topic

 Degree certificates and transcripts for Bachelor and Master (or the last 5 years)

 Master thesis (or equivalent) if it is already completed, or a description of the work in progress, otherwise

 The publications (or web links) of the candidate, if any (it is not expected that they have any)

 In addition, one recommendation letter from the person who supervises(d) the Master thesis (or research project or internship) should be sent directly by his/her author to the prospective PhD advisor.

Back

Top

Open fully-funded PhD position: “Hybrid Bayesian and deep neural modeling for weakly supervised

learning of sensory-motor speech representations”

The Deep-COSMO project, part of the new AI institute in Grenoble, is welcoming applications for a 3-year, fully funded

PhD scholarship starting October, 1st, 2019 at GIPSA-lab (Grenoble, France)

TOPIC: Representation learning, speech production and perception, Bayesian cognitive models, generative neural

networks

RESEARCH FIELD: Computer Science, Cognitive Science, Machine Learning, Artificial Intelligence, Speech Processing

SUPERVISION: J. Diard (LPNC); T. Hueber, L. Girin, J.-L. Schwartz (GIPSA-Lab)

IDEX PROJECT TITLE: Multidisciplinary Institute for Artificial Intelligence – Speech chair (P. Perrier)

SCIENTIFIC DEPARTMENT (LABORATORY’S NAME): GIPSA-lab

DOCTORAL SCHOOL: MSTII (maths and computer science) or EEATS (signal processing) or EDISCE (cognitive science),

depending on the candidate’s profile and career plan

TYPE of CONTRACT: 3-year doctoral contract

JOB STATUS: Full time

HOURS PER WEEK: 35

SALARY: between 1770 € and 2100 € gross per month (depending on complementary activity or not)

OFFER STARTING DATE: October, 1st, 2019

SUBJECT DESCRIPTION:

General objective

How can a child learn to speak from hearing sounds, without any motor instruction provided by his/her environment?

The general objective of this PhD project is to develop a computational agent, able to learn speech representations

from raw speech data in a weakly supervised configuration. This agent will involve an articulatory model of the human

vocal tract, an articulatory-to-acoustic synthesis system, and a learning architecture combining deep learning

algorithms and developmental principles inspired from cognitive sciences. This PhD will be part of the “Speech

communication” chair in the Multidisciplinary Institute for Artificial Intelligence in Grenoble (MIAI).

Method

This work will capitalize on two bricks of research recently developed in Grenoble. First, a Bayesian computational

model of speech communication called COSMO (Communicating about Objects using SensoriMotor Operations)

(Moulin-Frier et al., 2012, 2015; Laurent et al., 2017; Barnaud et al., 2019) was jointly developed by GIPSA and LPNC.

This model associates speech production and speech perception models in a single architecture. The random variables

in COSMO represent the signals and the sensori-motor processes involved in the speech production/perception loop.

COSMO learns their probability distributions from speech examples provided by the environment, and is then able to

perceive and produce speech sounds associated to speech categories. So far, COSMO was mostly tested on synthetic

data. One of the main challenges is now to confront COSMO to real-world data.

Second, we will also capitalize on a set of computational models for automatic processing and learning of sensorymotor

distributions in speech developed at GIPSA. This comprises a set of transfer-learning algorithms (Hueber et

al., 2015, Girin et al. 2017) aiming at adapting acoustic-articulatory knowledge on one speaker, towards another

speaker, using a limited amount of data, possibly incomplete and noisy; together with a set of deep neural networks

able to process raw articulatory data (Hueber et al., 2016; Tatulli & Hueber, 2017).

The first step will consist in designing, implementing and testing a “deep” version of COSMO, in which some of the

probability distributions are implemented by generative neural models (e.g. VAE, GAN). This choice is motivated by

the ability of such techniques to deal with raw, noisy and complex data, as well as their flexibility in terms of transfer

learning. The second stage will consist in reformulating entirely the speech communication agent in an end-to-end

neural architecture.

Outputs

The system will be tested in terms of both efficiency of the learning process – hence ability to generate realistic speech

sequences after convergence – and coherence of the motor strategies discovered by the computational agent, in spite

of the fact that no motor data will be provided for learning. The outputs are both (1) theoretical – for better

understanding the cognitive processes at hand in speech development and speech communication; (2) technical – for

integrating knowledge about speech production and cognitive processes in a machine learning architecture; and (3)

technological – for proposing a new generation of autonomous speech technologies.

ELIGIBILITY CRITERIA

Applicants must have:

- A Master's degree (or be about to earn one) or have a university degree equivalent to a European Master's (5-year

duration), in Computer Science, Cognitive Science, Signal Processing or Applied Mathematics.

- Solid skills in Machine Learning or probabilistic modeling + General knowledge in natural language processing

and/or speech processing (an affinity for cognitive sciences and speech sciences is welcome).

- Very good programming skills (mostly in Python).

- Good oral and written communication in English.

- Ability to work autonomously and in collaboration with supervisors and other team members.

SELECTION PROCESS

Applicants will have to send their CV + an application letter in English + copy of their last diploma to:

Jean-Luc.Schwartzr@gipsa-lab.fr, Thomas.Hueber@gipsa-lab.fr;

Letters of recommendation are welcome. Contact before preparing a complete application are welcome too.

Applications will be evaluated as they are received: the position is open until it is filled, with deadline on July 10th, 2019

Back

Top

PhD thesis proposal

Incremental sequence-to-sequence mapping for

speech generation using deep neural networks

June 17, 2019

1 Context and objectives

In recent years, deep neural networks have been widely used to address sequence-

to-sequence (S2S) learning. S2S models can solve many tasks where source

and target sequences have di
erent lengths such as: automatic speech recog-

nition, machine translation, speech translation, text-to-speech synthesis, etc.

Recurrent, convolutional and transformer architectures, coupled with attention

models, have shown their ability to capture and model complex temporal de-

pendencies between a source and a target sequence of multidimensional discrete

and/or continuous data. Importantly, end-to-end training alleviates the need

to previously extract handcrafted features from the data by learning hierarchi-

cal representations directly from raw data (e.g. character string, video, speech

waveform, etc.).

The most common models are composed of an encoder that reads the full in-

put sequence (i.e. from its beginning to its end) before the decoder produces the

corresponding output sequence. This implies a latency equals to the length of

the input sequence. In particular, for a text-to-speech (TTS) system, the speech

waveform is usually synthesized from a complete text utterance (e.g. a sequence

of words with explicit begin/end-of-utterance markers). Such approach cannot

be used in a truly interactive scenario, in particular by a speech-handicapped

person to communicate orally'. Indeed, the interlocutor has to wait for the

complete utterance to be typed before being able to listen to the synthetic voice,

hence limiting the dynamics and naturalness of the interaction.

The goal of this project is to develop a general methodology for incremental

sequence-to-sequence mapping, with application to interactive speech technolo-

gies. It will require the development of end-to-end classication and regression

neural models able to deliver chunks of output data on-the-y, from only a par-

tial observation of input data. The goal is to learn an ecient policy that leads

to an optimal trade-o
between (variable) latency and accuracy of the decoding

process. Possible strategies to decode the output data as soon as possible in-

clude: (i) Predicting online he future' of the output sequence from he past

1

and present' of the input sequence, with an acceptable tolerance to possible er-

rors, or (2) learn automatically from the data an optimal waiting policy' that

prevents the model to output data when the uncertainty is too high. The devel-

oped methodology will be applied to address two speech processing problems:

(i) Incremental Text-to-Speech synthesis in which speech is synthesized while

the user is typing the text (possibly with a variable latency), and (ii) Incremen-

tal speech enhancement/inpainting in which portions of the speech signal are

unintelligible because of sudden noise or speech production disorders, and must

be replaced on-the-y with reconstructed portions.

2 Work plan

The proposed working plan is the following :

 Bibliographic work on S2S neural models, in the context of speech recogni-

tion, speech synthesis, and machine translation as well as their incremental

(low-latency) variations

 Investigating new architectures, losses, and training strategies toward in-

cremental S2S models.

 Implementing and evaluating the proposed techniques in the context of

end-to-end neural TTS systems (the baseline system may be a neural

TTS trained with past information/left-context only).

 Implementing and evaluating the proposed techniques in the context of

speech enhancement/inpainting, rst on simulated noisy speech and then

on pathological speech.

3 Requirements

We are looking for an outstanding and highly motivated PhD candidate to work

on this subject. Following requirements are mandatory:

 Engineering degree and/or a Master's degree in Computer Science, Signal

Processing or Applied Mathematics.

 Solid skills in Machine Learning. General knowledge in natural language

processing and/or speech processing.

 Excellent programming skills (mostly in Python and deep learning frame-

works).

 Good oral and written communication in English.

 Ability to work autonomously and in collaboration with supervisors and

other team members.

2

4 Work context

Grenoble Alpes Univ. o
ers an excellent research environment with ample com-

puting facilities, as well as remarkable surroundings to explore over the week-

ends. The PhD project will be funded by the Grenoble Articial Intelligence

Institute (MIAI). The PhD candidate will work both at GIPSA-lab (CRISSP

team) and LIG-lab (GETALP team). The duration of the PhD is 3 years. The

salary is between 1770 and 2100 euros gross per month (depending on comple-

mentary activity or not).

5 How to apply?

Applications should include a detailed CV; a copy of their last diploma; at least

two references (people likely to be contacted); a cover letter of one page; a one-

page summary of the Master thesis; the two last transcripts of notes (Master or

engineering school). Applications should be sent to thomas.hueber@gipsa-lab.fr,

laurent.girin@gipsa-lab.fr and laurent.besacier@imag.fr. Applications will be

evaluated as they are received: the position is open until it is lled, with deadline

on July 10th, 2019.

Back

Top

 POST-DOC POSITION (18 months) - Forensic Voice Comparison (VoxCrim project): ability, limitations and specificities of listeners in speaker identification tasks
Laboratoire Parole et Langage (CNRS and Aix-Marseille Université) ? Aix-en-Provence, France

CONTEXT
The post-doc will be carried out within the framework of the VoxCrim project, funded by an ANR (Agence nationale de la recherche) grant (2017-2021, https://anr.fr/Project-ANR-17-CE39-0016). VoxCrim focuses on national security and legal/justice applications and aims to provide a validated scientific objective framework for all types of forensic voice comparison methods (automatic and phonetic). The goal is to develop certified standards to determine the specific areas in which voice comparison methods are applicable. The project includes two complementary subject areas: 1. the proposal of methodological standards to homogenize the expertise of voice comparison in a judicial environment, 2. the development of basic research in the fields of automatic speech processing and phonetics (speaker characteristics in the production and perception of speech).
The post-doc will participate in the second subject area (speaker characteristics in the production and perception of speech). Two questions need to be answered: What are the abilities and limits of listeners in speaker identification tasks? Which cues do listeners use to identify speakers?

TASKS
The main objective will be to conduct perception experiments aimed at assessing the ability of listeners in several speaker identification tasks.
The post-doctoral fellow will:
-    design experimental protocols
-    create and manipulate acoustic stimuli
-    run experiments and collect data
-    process data and perform statistical analysis
Finally, results will be presented at conferences and published in international journals.

WORK ENVIRONNEMENT
The postdoctoral fellow will work at the Laboratoire Parole et Langage (http://www.lpl-aix.fr/), a laboratory whose research interests are extremely varied (including linguistics, phonetics, neuroscience, psycholinguistics, sociolinguistics, and computer science). He or she will benefit from this stimulating environment and interact with all the members of the laboratory (faculty members, other post-docs, engineers, doctoral students, etc.). He or she will have the opportunity to discover all the projects of the laboratory.

EXPECTED PROFILE
The postdoctoral fellow will have a PhD in the speech sciences and/or in psychoacoustics (auditory measurements, audio signal processing). A strong background in data processing and statistics is also be required. A good command of French and English will also be appreciated.

18 months. Beginning in autumn 2019
Monthly salary: ~?1900  net (depending on experience)
Location : Laboratoire Parole et Langage (http://www.lpl-aix.fr/), Aix Marseille Université, CNRS UMR 7309, Aix-en-Provence, France
For additional information: christine.meunier@univ-amu.fr

-Application: CV and cover letter
-Send application as soon as possible to christine.meunier@univ-amu.fr

Supervisors: Dr. Christine Meunier, CNRS Researcher and Dr. Alain Ghio, Research Engineer - Laboratoire Parole et Langage, Aix-en-Provence, France.

Back

Top

RECRUTEMENT D?UN INGÉNIEUR D?ÉTUDES EN TRAITEMENT AUTOMATIQUE DES LANGUES NATURELLES ET EN DÉVELOPPEMENT D?UNE INTERFACE IHM-WEB

Début de contrat: Octobre 2019
Durée: 7 mois

Salaire: 2000? brut/mois


Profil :
- Titulaire d?un Master ou d?un doctorat en TAL
- Une formation en sciences du langage sera appréciée.
- Compétences opérationnelles en génie logiciel (gestion de version, tests, qualité de code) et Python

- Des compétences en C/C++ seraient un plus

- Une expérience en traitement automatique de la parole est requise ainsi qu?un bon niveau de français.

- Une experience en METEOR, firepad, Node.JS, mongoDB, firebase serait un plus.

Ce poste nécessite des capacités de travail en équipe et en autonomie.

Une connaissance du contexte linguistique de la surdité serait un atout supplémentaire.

*Description du projet et des missions*
Dans le cadre du projet MANES (Médiation et Accessibilité Numérique pour les Étudiants Sourds) dirigé par François Portet (LIG), Isabelle Estève (LIDILEM) et Marion Fabre (ECP) dont une partie du financement est assurée par PULSALYS, IDEX Lyon-Saint Etienne, nous recrutons un ingénieur d?études pour un CDD de 7 mois.

L?objectif général du projet est de développer un dispositif de sous-titrage en temps-réel pour rendre accessible le discours oral de l?enseignant aux étudiants sourds, de façon à favoriser l?appropriation individuelle des savoirs, par le biais de la prise de notes. La réalisation technologique et les capacités de traitement de l?écrit par les publics sourds seront les deux axes de ce projet.

La mission de l?ingénieur d?études consiste d?une part à développer, évaluer et améliorer des prototypes basés sur les dernières avancées scientifiques et à les fusionner pour réaliser un prototype de sous-titrage automatique en temps réel, à partir de la plateforme Kaldi.

D?autre part, à concevoir une interface IHM pour la retranscription automatique en temps réel du discours de l?enseignant et la projection du sous-titrage en cours, en y intégrant le prototype mentionné ci-dessus.

Le candidat aura en charge l?élaboration du prototype pour des expérimentations en salle de cours et les réajustements de l?interface liés à ces expérimentations.

Missions en Traitement automatique des Langues :
- Prise de connaissance de l?état de l'art des systèmes de sous-titrage automatique
- Test de transcription semi-automatique et vérification des extraits oraux de cours magistraux
- Adaptation du système temps-réel de transcription automatique à réseaux de neurones (Kaldi)
- Traitement en temps-réel des transcriptions pour le sous-titrage adapté aux publics sourds

Les exigences fonctionnelles envisagées pour l?implémentation de la plateforme Kaldi sont : le repérage des mots-clés et des synonymes en temps réel (déjà existant dans KALDI) et le développement de nouvelles fonctionnalités : la segmentation et la simplification. Des perspectives d?adaptations pour le Off-Line seront aussi à envisager.


Missions en développement :
- Réalisation d?une application web permettant la retranscription en temps réel du discours de l?enseignant et la projection de la retranscription obtenue (vidéo projecteur et possibilité d?extension pour une interface mobile)
- Elaboration de l?interface étudiant : stockage, récupération de la trace écrite pour retravail et modification.
- Elaboration de l?interface enseignant : paramétrisation des éléments clés du cours
- Documentation : description et mode d?emploi de l?interface d?IHM.


*Environnement de travail*

Le projet est porté par le laboratoire Education, Cultures et Politiques (ECP, EA 4571), Université Lumière Lyon 2 encollaboration avec le Laboratoire de Linguistique et Didactique des Langues Étrangères et Maternelles (LIDILEM), Université Grenoble-Alpes et le Laboratoire d'Informatique de Grenoble (LIG).

Le poste sera accueilli physiquement au Laboratoire d'Informatique de Grenoble, UMR CNRS, au sein de l'équipe GETALP. L'équipe GETALP (https://lig-getalp.imag.fr) regroupe plus de 40 chercheurs, ingénieurs et étudiants dans le domaine du traitement automatisé des langues et de la parole multilingue.

*Candidature*
Envoyer un CV, une lettre de motivation et 1 à 3 lettres de recommandation à Marion.Fabre@univ-lyon2.fr, François.portet@imag.fr, Benjamin.Lecouteux@imag.fr et isabelle.esteve@univ-grenoble-alpes.fr.
Les candidatures seront examinées dès réception, la convocation éventuelle à un entretien d?embauche est prévue début juillet. Merci de candidater dès que possible et avant le 1^er Juillet 2019 minuit.

Back

Top

Post-doc position open
------------------------------------
The Speech and Language Technology Group in Le Mans University is looking for
a post-doc scientist to develop autonomous systems

Keywords: Deep Learning, lifelong, autonomous systems, unsupervised learning,
active-learning, interactive-learning

Context
------------------------------------
The LST team from LIUM (Le Mans University) is focusing on autonomous system?s  behavior
for the task of speaker diarization and machine translation. The ALLIES project
(European Chist-ERA collaborative project) aims at developing evaluation protocols,
metrics
and scenarios for lifelong learning autonomous systems. The goal is to enable
auto-adaptable
systems that can also auto-evaluate in order to sustain their performance across time.
Autonomous systems can rely on human domain experts via active and interactive learning
processes to be define within the ALLIES project.

Missions
------------------------------------
Develop an autonomous system for speaker diarization by integrating lifelong learning,
active and interactive learning components. The research work will be related to some of
the following topics:
    - unsupervised adaptation
    - unsupervised evaluation
    - active learning (based on the unsupervised evaluation process, the autonomous system
        is free to require additional knowledge from the human domain expert)
    - Interactive learning (a human domain expert provides specific knowledge to the
autonomous
        system.  This  information must be taken into account by the system)
    - Performance will be analyzed using protocols, metrics and scenarios developed for
the ALLIES project.

Participation to the ALLIES benchmarking evaluation for speaker diarization.
During the ALLIES project, LIUM is organizing two international evaluation
campaigns (one for Speaker Diarization jointly organized with Albayzin and the
second one for Machine Translation jointly with WMT)
The benchmarking evaluation will serve to validate approaches developed during the
post-doc

Dissemination
The research will be published in the major conferences and journals

------------------------------------
Duration: 12 months
Salary: 2 365,14? (after taxes)

Start: as soon as possible, latest January 2020

Location: LIUM, Le Mans University

Supervisers: Anthony Larcher (anthony.larcher@univ-lemans.fr) and Loïc Barrault
(loic.barrault@univ-lemans.fr)

Expected competences:
    - Phd in Machine Learning and Deep Learning
    - Experience in speech processing is positive
    - Python fluent
    - familiar with a deep learning toolkit (Pytorch, TensorFlow)

ALLIES website: https://projets-lium.univ-lemans.fr/allies/

Back

Top

ZAION est une société innovante en pleine expansion spécialisée dans la technologie des robots conversationnels : callbot et chatbot intégrant de l’Intelligence Artificielle.

ZAION a développé une solution qui s’appuie sur une expérience de plus de 20 ans de la Relation Client. Cette solution en rupture technologique reçoit un accueil très favorable au niveau international et nous comptons déjà 18 clients actifs (GENERALI, MNH, APRIL, CROUS, EUROP ASSISTANCE, PRO BTP …).

Nous sommes actuellement parmi les seuls au monde à proposer une offre de ce type entièrement tournée vers la performance. Nous rejoindre, c’est prendre part à une aventure passionnante au sein d’une équipe ambitieuse afin de devenir la référence sur le marché des robots conversationnels.

Dans le cadre de son développement ZAION recherche son Responsable de IA H/F. Manager de l’équipe R&D, votre rôle est stratégique dans le développement et l’expansion de la société. Vous développerez, une solution qui permet de détecter les émotions dans les conversations. Nous souhaitons augmenter les fonctionnalités cognitives de nos callbots afin qu’ils puissent détecter les émotions de leurs interlocuteurs (joie, stress, colère, tristesse…) et donc adapter leurs réponses en conséquence.

Vos missions principales :

- Vous participez à la création du pôle R&D de ZAION et piloterez à votre arrivée votre premier projet de reconnaissance d’émotion dans la voix.

- Construisez, adaptez et faites évoluer nos services de détection d’émotion dans la voix

- Analysez de bases de données conséquentes de conversations pour en extraire les conversations émotionnellement pertinentes

- Construisez une base de données de conversations labelisées avec des étiquettes émotionnelles

- Formez et évaluez des modèles d'apprentissage automatique pour la classification d’émotion

- Déployez vos modèles en production

- Améliorez en continue le système de détection des émotions dans la voix

Qualifications requises et expérience antérieure :

-Vous avez une expérience de 5 ans minimum comme Data Scientist/Machine Learning appliqué à l’Audio et une appétence à l’encadrement 

- Diplômé d’une école d’Ingénieur ou Master en informatique ou un doctorat en informatique mathématiques avec des compétences solides en traitements de signal (audio de préférence)

- Solide formation théorique en apprentissage machine et dans les domaines mathématiques pertinents (clustering, classification, factorisation matricielle, inférence bayésienne, deep learning...)

- La mise à disposition de modèles d'apprentissage machine dans un environnement de production serait un plus

- Vous maîtrisez un ou plusieurs des langages suivants : Python, Frameworks de machine Learning/Deep Learning (Pytorch, TensorFlow,Sci-kit learn, Keras) et Javascript

- Vous maîtrisez les techniques du traitement du signal audio

- Une expérience confirmée dans la labélisation de grande BDD (audio de préférence) est indispensable ;

- Votre personnalité : Leader, autonome, passionné par votre métier, vous savez animer une équipe en mode projet

- Vous parlez anglais couramment

Merci d’envoyer votre candidature à : alegentil@zaion.ai

Back

Top

1.SPEECH RECOGNITION INTERN

Location: Cambridge, UK

Contact: careers@speechmatics.com

“As an intern at Speechmatics I have worked on projects that use real machine learning to deliver real value to people across the world. There are few places where the machine learning being used is at the bleeding edge of the field, but Speechmatics is one of them. The company has an amazing culture that allows you to grow as a programmer and as a person. If you want to be a part of a fast-growing machine learning company where you, personally, will make a difference then Speechmatics could well be the place for you!” 

• Sam Ringer, Machine Learning Engineer (previously R&D Intern), Speechmatics 

Background

Speech technology is one of the most popular discussion items at the moment, yet speech interaction is limited to “Alexa, turn on the light”, or “Siri, where is the nearest coffee shop?” We are taking speech technology to the next level using our expertise in machine learning and speech-to-text technology to enable our customers to use conversational speech recognition. Our solutions power subtitling on TV, content discovery for videos, compliance solutions in banks, improve efficiency of meetings, and many other use-cases. Our mission is to improve human communication with a global speech engine, that works and put speech back at the heart of communication.

At Speechmatics you’ll be working with some of the smartest minds in the industry, working on cutting-edge projects and deploying the latest machine learning techniques to disrupt the market, providing customers with the best speech technology available, all whilst immersed in a progressive and great company culture. You can enjoy benefits including, share options, healthcare, life assurance, Bike Doctor, massages, regular BBQs, Brew Dogs in the fridge, no red tape, a top end laptop and much more. We’re building a company that truly strives to be world-leading and we’re looking for people who wholeheartedly believe they can be additive to our culture, bring new ideas to the table and get stuff done. If that’s you, carry on reading.

The Opportunity

The Speechmatics Engineering team develops and maintains speech-oriented products and services that will be used by businesses worldwide and is responsible for the complete product development cycle for these products. In this internship, you’ll help to support fundamental speech and language processing research to improve our performance and language coverage as well as helping to build products and features to delight our users.

Because you will be joining a rapidly expanding team, you will need to be a team player who thrives in a fast-paced environment, with a focus on investigating ideas and rapidly moving research developments into products. We strongly encourage versatility and knowledge transfer within and across teams. You will be expected to learn fast and feel emboldened to ask for support as you need it.

Prior experience of speech recognition is desirable, although Speechmatics has a team of speech recognition engineers who will collaborate and share any specialised knowledge required. If you are enthusiastic about speech recognition and machine learning in general, with the drive to deliver the best possible technology solutions, then we want to hear from you!

Our internships are not time constrained to specific dates – we can work out mutually agreeable start and end dates as part of the application process.

Key Responsibilities

• Exploring and evaluating research ideas

• Increasing and improving our language coverage

• Prototyping new and improved features

• Helping the company to take your R&D through to production

• Communicating your work internally

Requirements

Essential

• Team player

• Enthusiasm for speech recognition and machine learning

• Technical understanding of speech recognition or related discipline

• Ability to rapidly deliver on ideas

• Competent in Python and/or C/C++

• Have or be studying towards a degree involving speech recognition, machine learning / computer science or related field

Desirable

• Practical experience of ASR and ML packages such as Kaldi, HTK or TensorFlow

• Commercial experience of speech recognition

• Software development experience

Salary

Competitive salary (dependent on experience), flexible working and some awesome benefits & perks.

Interested?

Get in touch! Send your CV and covering letter to careers@speechmatics.com.

2.SPEECH RECOGNITION ENGINEER

Location: Cambridge, UK

Contact: careers@speechmatics.com

'As a Speech Recognition Engineer at Speechmatics, I work on solving a multitude of problems related to improving the accuracy and delivering new features for a global automatic speech recognition engine. As a member of the speech team, I work across every aspect of speech and implement the latest research in acoustic and language modelling. The team is supportive and also rich in terms of skills and backgrounds. Speechmatics offer progressive and rewarding opportunities in one of the best speech technology companies in the world.' 

• André Mansikkaniemi, Speech Recognition Engineer at Speechmatics 

Background

Speech technology is one of the most popular discussion items at the moment, yet speech interaction is limited to “Alexa, turn on the light”, or “Siri, where is the nearest coffee shop?” We are taking speech technology to the next level using our expertise in machine learning and speech-to-text technology to enable our customers to use conversational speech recognition. Our solutions power subtitling on TV, content discovery for videos, compliance solutions in banks, improve efficiency of meetings, and many other use-cases. Our mission is to improve human communication with a global speech engine, that works and put speech back at the heart of communication.

At Speechmatics you’ll be working with some of the smartest minds in the industry, working on cutting-edge projects and deploying the latest machine learning techniques to disrupt the market, providing customers with the best speech technology available, all whilst immersed in a progressive and great company culture. You can enjoy benefits including, share options, healthcare, life assurance, Bike Doctor, massages, regular BBQs, Brew Dogs in the fridge, no red tape, a top end laptop and much more. We’re building a company that truly strives to be world-leading and we’re looking for people who wholeheartedly believe they can be additive to our culture, bring new ideas to the table and get stuff done. If that’s you, carry on reading.

The Opportunity 

We are looking for a talented speech engineer to help us build the best speech technology for anybody, anywhere, in any language. You will be part of a team that is working on our core ASR capabilities to improve our speed and accuracy and develop novel features that we can support in all languages. Your work will feed into our ground-breaking framework to support the building of ASR models in every language pack published by the company. You will be responsible for keeping our system the most accurate and useful commercial speech recognition system available. 

As you will be joining a small team, you will need to be a team player who thrives in a fast-paced environment, with a focus on rapidly moving research developments into products. Bringing skills into the team is as important as a can-do attitude. We strongly encourage versatility and knowledge transfer within the team, so that we can share efficiently what needs to be done to meet our commitments to the rest of the company. 

Key Responsibilities 

• Research and development of improved speed and accuracy across our range of world leading ASR products and related features

• Delivering the software that provides an easy-to-use feature rich ASR product for our customers

• Enhancing our machine learning framework that robustly builds any language with the best possible performance

• Taking data all the way from its raw form through to a finished model

• Working within a team in an agile environment

• Working closely with other technical teams and product team to deliver on the company’s technical vision

Requirements 

Essential 

• Graduate degree in Statistics, Engineering, Mathematics, Computer Science 

• Knowledge of key natural language processing or related technologies, such as speech recognition, text-to-speech or natural language understanding 

• Experience working with standard speech and/or ML toolkits, e.g. Kaldi, KenLM, TensorFlow, etc. 

• Solid Python programming skills

• Experience using Unix/Linux 

• Quick and enthusiastic learner

• Excellent teamwork and communications skills

• Analytical mind-set with a data-driven approach to making decisions and attention to detail 

Desirable 

• Postgraduate degree in related discipline 

• Commercial work experience in ASR or a related field 

• Experience of working in an Agile framework 

• Expertise in modern speech recognition, including WFSTs, lattice processing, neural net (RNN / DNN / LSTM), acoustic and language models, decoding 

• Comprehensive knowledge of machine learning and statistical modelling 

• Experience in deep machine learning and related toolkits, e.g. Theano, Torch, etc. 

• Deep expertise in Python and/or C++ software development 

• Experience working effectively with software engineering teams or as a Software Engineer 

Salary

Competitive salary (dependent on experience), flexible working and some awesome benefits & perks.

Interested?

Get in touch! Send your CV and covering letter to careers@speechmatics.com.

3.SENIOR SPEECH RECOGNITION ENGINEER

Location: Cambridge, UK

Contact: careers@speechmatics.com

'As a Speech Recognition Engineer at Speechmatics, I work on solving a multitude of problems related to improving the accuracy and delivering new features for a global Automatic Speech Recognition engine. As a member of the speech team, I work across every aspect of speech and implement the latest research in acoustic and language modelling. The team is supportive and also rich in terms of skills and backgrounds. Speechmatics offer progressive and rewarding opportunities in one of the best speech technology companies in the world.'

• André Mansikkaniemi, Speech Recognition Engineer, Speechmatics

Background

Speech technology is one of the most popular discussion items at the moment, yet speech interaction is limited to “Alexa, turn on the light”, or “Siri, where is the nearest coffee shop?” We are taking speech technology to the next level using our expertise in machine learning and speech-to-text technology to enable our customers to use conversational speech recognition. Our solutions power subtitling on TV, content discovery for videos, compliance solutions in banks, improve efficiency of meetings, and many other use-cases. Our mission is to improve human communication with a global speech engine, that works and put speech back at the heart of communication.

At Speechmatics you’ll be working with some of the smartest minds in the industry, working on cutting-edge projects and deploying the latest machine learning techniques to disrupt the market, providing customers with the best speech technology available, all whilst immersed in a progressive and great company culture. You can enjoy benefits including, share options, healthcare, life assurance, Bike Doctor, massages, regular BBQs, Brew Dogs in the fridge, no red tape, a top end laptop and much more. We’re building a company that truly strives to be world-leading and we’re looking for people who wholeheartedly believe they can be additive to our culture, bring new ideas to the table and get stuff done. If that’s you, carry on reading.

The Opportunity

We are looking for a talented speech engineer to help us build the best speech technology for anybody, anywhere, in any language. You will be part of a team that is working on our core ASR capabilities to improve our speed and accuracy and develop novel features that we can support in all languages. Your work will feed into our ground-breaking framework to support the building of ASR models in every language pack published by the company. You will be responsible for keeping our system the most accurate and useful commercial speech recognition system available.

As you will be joining a small team, you will need to be a team player who thrives in a fast-paced environment, with a focus on rapidly moving research developments into products. Bringing skills into the team is as important as a can-do attitude. We strongly encourage versatility and knowledge transfer within the team, so that we can share efficiently what needs to be done to meet our commitments to the rest of the company.

Key Responsibilities

• Research and development of improved speed and accuracy across our range of world leading ASR products and related features

• Delivering the software that provides an easy-to-use feature rich ASR product for our customers

• Enhancing our machine learning framework that robustly builds any language with the best possible performance

• Taking data all the way from its raw form through to a finished model

• Working within a team in an agile environment

• Working closely with other technical teams and product team to deliver on the company’s technical vision

Requirements

Essential

• Commercial experience in ASR or a related field

• Graduate degree in Statistics, Engineering, Mathematics, or Computer Science

• Expertise in modern speech recognition, including WFSTs, lattice processing, neural net (RNN / DNN / LSTM), acoustic and language models, decoding

• Experience working with standard speech and/or ML toolkits, e.g. Kaldi, KenLM, TensorFlow, etc.

• Solid Python programming skills

• Experience using Unix/Linux

• Drive to help those around you learn and improve every day

• Excellent teamwork and communications skills

• Analytical mind-set with a data-driven approach to making decisions and attention to detail

Desirable

• Postgraduate degree in related discipline

• Experience of working in an Agile framework

• Comprehensive knowledge of machine learning and statistical modelling

• Experience in deep machine learning and related toolkits, e.g. Theano, Torch, etc.

• Deep expertise in Python and/or C++ software development

• Experience working effectively with software engineering teams or as a Software Engineer

Salary

Competitive salary (dependent on experience), flexible working and some awesome benefits & perks.

Interested?

Get in touch! Send your CV and covering letter to careers@speechmatics.com.

More about Speechmatics’ culture

Live for the wow | Build authentic relationships | Be the adventure

Innovation is what we do. We build, we iterate, we develop the next thing that delivers that wow moment. We see value in building long-term, authentic relationships that last and are based on trust and honesty. With our customers, our colleagues, our leaders, our suppliers or within our local community. Our journey should be fun and exciting. We will celebrate our successes and learn from our mistakes together along the way. We embrace learning and change to grow naturally and organically as a company and individuals. We trust, we’re honest, kind and respectful.

Back

Top

Applications are invited for a three-year Early Stage Researcher PhD positions in the

speech technology for pathological speech.

Description

The thesis focuses on studying the link between the internal representations of Deep Neural Networks (DNNs) and

the subjective representation of speech intelligibility. We propose to explore the saliency detection capabilities of

DNNs when used in a regression task for predicting speech intelligibility scores as given by human experts. By

saliency, we mean to retrieve which frequency bands are important and used by a DNN to make its predictions.

The final expectation is to identify regions of interest in the speech signal, both in time and frequency, that

characterise the level of speech impairment.

The experiments will be processed on various samples of speech performed by 150 people (100 patients and 50

healthy controls). This database was recorded within the INCA C2SI project, and contains speech from patients

treated for cancer of the oral cavity or pharynx. It contains also various metadata such as the location of the tumor,

the impairment in terms of severity and intelligibility that were appreciated by human experts, self evaluation

questionnaires on the patient’s quality of life… Various tasks were recorded such as a sustained vowel, read

speech, nonsense words, prosodic exercises, picture description, etc. There will be also the possibility to extend

the work to another corpus which is composed of voice of patients suffering from Parkinson disease.

At first, the PhD will have to take benefit from the various analysis and descriptions that were done during the C2SI

project trying to correlate the impact of the tumor and the communication ability. Those results will help attesting

the human representation of the impact of the disease. Then, a DNN representation will be modeled to fit the data,

taking care of the data sparsity. The last part of the work will be to explore the intern representation of the DNN,

trying to explore what part of the signal help to make a decision on the impact of the disease and that will be the

final goal of the thesis, studying the automatic representation that lies in the model the student will propose.

This work is funded by the TAPAS project (https://www.tapas-etn-eu.org) which is a Horizon 2020 Marie

Skłodowska-Curie Actions Initial Training Network European Training Network (MSCA-ITN-ETN) project that aims

to transform the well being of people across Europe with debilitating speech pathologies (e.g., due to stroke,

Parkinson's, etc.). These groups face communication problems that can lead to social exclusion. They are now

being further marginalised by a new wave of speech technology that is increasingly woven into everyday life but

which is not robust to atypical speech.

The supervision of the PhD will take place at IRIT laboratory by the SAMoVA team in Toulouse. SAMoVA does

research in the domain of “analysis, modeling and structuring of audiovisual content”. The application areas are

diverse: speech processing, identification of languages, speaker verification and speech and music indexing. The

researchers expertise covers novel machine learning and audio processing technologies and is now focused on

deep learning methods, leading to several publications in international conferences.

Eligibility Criteria

Early Stage Researchers (ESRs) shall, at the time of recruitment by the host organization, be in the first four

years (full-time equivalent research experience) of their research careers.

- The ESR may be a national of a Member State, of an Associated Country or of any Third Country.

- The ESR must not have resided or carried out her/his main activity (work, studies, etc.) in the country of her/his

host organization for more than 12 months in the 3 years immediately prior to her/his recruitment.

- Holds a Master’s degree or equivalent, which formally entitles to embark on a Doctorate.

- Does not hold a PhD degree.

Duration of recruitement: 36 months

Contact: Julie Mauclair (mauclair@irit.fr)

Back

Top

Chief Technical Officer (CTO)

Under the supervision of the CEO, the responsibilities of the Chief Technical Officer (CTO) include planning and supervising technical development of tools, software components or applications for language resource production and management.
He/she will be in charge of managing the current language resources production workflows and co-ordinating ELDA?s participation in R&D projects while being also hands-on whenever required by the language resource production and management team. He/she will liaise with external partners at all phases of the projects (submission to calls for proposals, building and management of project teams) within the framework of international, publicly- or privately-funded projects.

This yields excellent opportunities for creative and motivated candidates wishing to participate actively to the Language Engineering field.

Profile:
?    PhD in Computer Science, Natural Language Processing, or equivalent
?    Experience in Natural Language Processing (speech processing, data mining, machine translation, etc.)
?    Familiarity with open source and free software
?    Knowledge of a statically typed functional programming language (OCaml preferred) is a plus
?    Good level in English, with strong writing and documentation skills in English
?    Dynamic and communicative, flexible to work on different tasks in parallel
?    Ability to work independently and as part of a multidisciplinary team
?    Citizenship (or residency papers) of a European Union country
?    Good level in Python, knowledge of Django would be a plus
?    Proficiency in classic shell scripting in a Linux environment (POSIX tools, Bash, awk)

Salary: Commensurate with qualifications and experience (between 45-55K?).
Other benefits: complementary health insurance and meal vouchers

Applicants should email a cover letter addressing the points listed above together with a curriculum vitae to: job@elda.org

ELDA is acting as the distribution agency of the European Language Resources Association (ELRA). ELRA was established in February 1995, with the support of the European Commission, to promote the development and exploitation of Language Resources (LRs). Language Resources include all data necessary for language engineering, such as monolingual and multilingual lexica, text corpora, speech databases and terminology. The role of this non-profit membership Association is to promote the production of LRs, to collect and to validate them and, foremost, make them available to users. The association also gathers information on market needs and trends.

For further information about ELDA/ELRA, visit: ww.elra.info

Back

Top

 2 Post-doctoral positions at Le Mans University on Deep learning approaches speech processing

*Place of work* Le Mans University, Le Mans ? France

*Starting date* From now to June 2020

*Salary* between 2 300 and 2 600 ? /month

*Duration* 12 months and 24 months (can be combined in a 36 months position)

****************************************
1st position
****************************************

* Context *
The LST team from LIUM (Le Mans University) is focusing on autonomous system?s  behavior
for the task of speaker diarization and machine translation.
The ALLIES project (European Chist-ERA collaborative project) aims at developing
evaluation protocols, metrics and scenarios for lifelong learning autonomous systems.
The goal is to enable auto-adaptable systems that can also auto-evaluate in order to
sustain their performance across time. Autonomous systems can rely on human domain
experts via active and interactive learning processes to be define within the ALLIES project.

* Missions *
Develop an autonomous system for speaker diarization by integrating lifelong learning,
active and interactive learning components. The research work will be related to some of the following topics:
- unsupervised adaptation
- unsupervised evaluation
- active learning (based on the unsupervised evaluation process, the autonomous
   system is free to require additional knowledge from the human domain expert)
- Interactive learning (a human domain expert provides specific knowledge to
   the autonomous system.  This  information must be taken into account by the system)
Performance will be analyzed using protocols, metrics and scenarios developed for the ALLIES project.

Participation to the ALLIES benchmarking evaluation for speaker diarization.
During the ALLIES project, LIUM is organizing two international evaluation
campaigns (one for Speaker Diarization jointly organized with Albayzin and the
second one for Machine Translation jointly with WMT)
The benchmarking evaluation will serve to validate approaches developed during the post-doc

* Dissemination*
The research will be published in the major conferences and journals

* Duration * 12 months
* Salary * 2 365,14? (after taxes)

* Start * as soon as possible, latest January 2020

* Supervisers * Anthony Larcher (anthony.larcher@univ-lemans.fr) and Loïc Barrault (loic.barrault@univ-lemans.fr)

Expected competences:
    - Phd in Machine Learning and Deep Learning
    - Experience in speech processing is positive
    - Python fluent
    - familiar with a deep learning toolkit (Pytorch, TensorFlow)

ALLIES website: https://projets-lium.univ-lemans.fr/allies/

****************************************
2nd position
****************************************

* Context *
The LST team from LIUM (Le Mans University) is focusing on evolutive end-to-end
neural networks for speaker recognition. The Extensor project (French ANR funded)
aims at developing novel architectures for end-to-end speaker recognition as well as
explaining the behavior of those networks. The focus of Extensor is threefold:
get rid of the legacy of bayesian system?s architecture and explore wider opportunities offered in deep learning;
explore real end-to-end architectures exploiting the tax signal instead of classical features (such as MFCC of filterbanks);
Develop tools for explainability in speaker recognition.

* Missions *
Develop end-to-end speaker recognition system based on state-of-the-art approaches  (x-vectors, sincnet?)
Develop evolutive architectures making use of existing genetic algorithms and study their behavior.
Participate to the three hackathons organized by the Extensor project in order to develop
tools for evolutive neural network architecture and explainability for speaker recognition.
Dissemination: the research will be published in the major conferences and journals

* Duration * 24 months
* Salary * 2 600? (after taxes)

* Start * as soon as possible, latest June 2020

* Location * LIUM, Le Mans University

* Superviser * Anthony Larcher (anthony.larcher@univ-lemans.fr)

Expected competences:
    - Phd in Machine Learning and Deep Learning
    - Experience in speech processing is positive
    - Python fluent
    - familiar with a deep learning toolkit (Pytorch, TensorFlow)

Back

Top

Applications are invited for a three-year Early Stage Researcher PhD positions in the

speech technology for pathological speech.

Description

The thesis focuses on studying the link between the internal representations of Deep Neural Networks (DNNs) and

the subjective representation of speech intelligibility. We propose to explore the saliency detection capabilities of

DNNs when used in a regression task for predicting speech intelligibility scores as given by human experts. By

saliency, we mean to retrieve which frequency bands are important and used by a DNN to make its predictions.

The final expectation is to identify regions of interest in the speech signal, both in time and frequency, that

characterise the level of speech impairment.

The experiments will be processed on various samples of speech performed by 150 people (100 patients and 50

healthy controls). This database was recorded within the INCA C2SI project, and contains speech from patients

treated for cancer of the oral cavity or pharynx. It contains also various metadata such as the location of the tumor,

the impairment in terms of severity and intelligibility that were appreciated by human experts, self evaluation

questionnaires on the patient’s quality of life… Various tasks were recorded such as a sustained vowel, read

speech, nonsense words, prosodic exercises, picture description, etc. There will be also the possibility to extend

the work to another corpus which is composed of voice of patients suffering from Parkinson disease.

At first, the PhD will have to take benefit from the various analysis and descriptions that were done during the C2SI

project trying to correlate the impact of the tumor and the communication ability. Those results will help attesting

the human representation of the impact of the disease. Then, a DNN representation will be modeled to fit the data,

taking care of the data sparsity. The last part of the work will be to explore the intern representation of the DNN,

trying to explore what part of the signal help to make a decision on the impact of the disease and that will be the

final goal of the thesis, studying the automatic representation that lies in the model the student will propose.

This work is funded by the TAPAS project (https://www.tapas-etn-eu.org) which is a Horizon 2020 Marie

Skłodowska-Curie Actions Initial Training Network European Training Network (MSCA-ITN-ETN) project that aims

to transform the well being of people across Europe with debilitating speech pathologies (e.g., due to stroke,

Parkinson's, etc.). These groups face communication problems that can lead to social exclusion. They are now

being further marginalised by a new wave of speech technology that is increasingly woven into everyday life but

which is not robust to atypical speech.

The supervision of the PhD will take place at IRIT laboratory by the SAMoVA team in Toulouse. SAMoVA does

research in the domain of “analysis, modeling and structuring of audiovisual content”. The application areas are

diverse: speech processing, identification of languages, speaker verification and speech and music indexing. The

researchers expertise covers novel machine learning and audio processing technologies and is now focused on

deep learning methods, leading to several publications in international conferences.

Eligibility Criteria

Early Stage Researchers (ESRs) shall, at the time of recruitment by the host organization, be in the first four

years (full-time equivalent research experience) of their research careers.

- The ESR may be a national of a Member State, of an Associated Country or of any Third Country.

- The ESR must not have resided or carried out her/his main activity (work, studies, etc.) in the country of her/his

host organization for more than 12 months in the 3 years immediately prior to her/his recruitment.

- Holds a Master’s degree or equivalent, which formally entitles to embark on a Doctorate.

- Does not hold a PhD degree.

Duration of recruitment: 36 months.

Contact : Julie Mauclair (mauclair@irit.fr)

Back

Top

PhD position in

Agent Based Modeling of

Cognitively Plausible Emergent Behavior

In the context of seed funding for AI research in Flanders prof. Bart de Boer is looking for a PhD student for the origins of language group of the AI-lab of the Vrije Universiteit Brussels.

PhD position offered

We offer a four year PhD position funded by a scholarship with a yearly bench fee. The PhD work will consist of building an agent-based simulation in which we can investigate emergence of behavior in a cognitively realistic setting. This means that the agents are not fully rational and that they show behavior similar to that of humans, and that interests of agents are not necessarily always aligned. The modeling will primarily focus on emergence of speech, but the simulation should be general enough that it can be easily adapted to other areas, such as traffic or economic interactions.

What we are looking for

We are looking for an enthusiastic student with a degree in artificial intelligence, cognitive science, linguistics or equivalent and who has experience programming agent-based or cognitive models, preferably in Python or C++. Knowledge of speech and speech processing is a bonus. The starting date is negotiable, but preferably no later than September 2019.

How to apply

Send a recent CV, detailing your academic record and your programming experience as well as a letter of motivation to prof. Bart de Boer. At this stage we ask you not to send copies of your diplomas or letters of reference. These we will request directly if we decide to further pursue your application, If you have any questions please email prof. Bart de Boer.

Links

Context: https://ai.vub.ac.be/node/1687

Email Bart de Boer: bart@ai.vub.ac.be

Back

Top

Visiting postdoc in

Cognitively Plausible Emergent Behavior

In the context of seed funding for AI research in Flanders prof. Bart de Boer is looking for a short term (three-six months) visiting postdoc for the origins of language group of the AI-lab of the Vrije Universiteit Brussels.

Position offered

We offer a three-six months visiting postdoc position funded by a scholarship and with a bench fee. The work should consist of agent-based simulation, or of experiments to investigate emergence of behavior in a cognitively realistic setting. This means that in a computer simulation, the agents are not fully rational and that they show behavior similar to that of humans, and that interests of agents are not necessarily always aligned. Experiments should focus on factors that are typical for human settings, but that are generally idealized away, such as altruism, conflicts of interests and other 'non-rational' behaviors. We are most interested in modeling emergence of speech, but we welcome applications proposing other areas, such as traffic or economic interactions.

What we are looking for

We are looking for an enthusiastic postdoc with a track record in artificial intelligence, cognitive science, linguistics or equivalent and who has either experience programming agent-based or cognitive models, or who has experience with the interaction between computer models and experiments. The starting date is negotiable, but preferably no later than September 2019.

How to apply

Send a recent CV, detailing your academic record and your programming experience as well as a letter of motivation to prof. Bart de Boer. Be sure to include a short (1-page) outline of your proposed project in the letter of motivation, as well as a short planning. At this stage we ask you not to send copies of your diplomas or letters of reference. These we will request directly if we decide to further pursue your application, If you have any questions please email prof. Bart de Boer.

Links

Context: https://ai.vub.ac.be/node/1687

Email Bart de Boer: bart@ai.vub.ac.be

Back

Top

EURECOM (Nice, France), Inria (Nancy, France) and LIA (Avignon, France) are opening a
18-month Research Engineer or Postdoc position on speaker de-identification and voice
privacy.

For more information and to apply:
https://jobs.inria.fr/public/classic/en/offres/2019-01937

Back

Top

Ph.D. position in Softbank robotics and Telecom-Paris
 
Subject:  Automatic multimodal recognition of users? social behaviors
in human-robot interactions (HRI)

*Places of work*  Softbank Robotics [SB] (Paris 15e) & Telecom Paris [TP] Palaiseau (Paris outskirt)

*Starting date* December 2019

*Funding* CIFRE http://www.anrt.asso.fr/fr/cifre-7843

*Context*
The research activity of the Ph.D. candidate will contribute to :
- Softbank Robotics robot?s software NAOqi, within the Expressivity team responsible for ensuring an expressive, natural and fun interaction with our robots.
- the Social Computing topic [SocComp.] of the S2a team [SSA] at Telecom-ParisTech, in close collaboration with other researchers and Ph.D. students of the team.

* Candidate profile*
As a minimum requirement, the successful candidate should have:
?    A master in one or more of the following areas: human-agent interaction, deep learning, computational linguistics, cognitive sciences, affective computing, reinforcement learning, natural language processing, speech processing
?    Excellent programming skills (preferably in Python)
?    Excellent command of English
?    Very good communication skills, commitment, independent working style as well as initiative and team spirit

Given the multidisciplinary aspect of the subject, priority will be given to multidisciplinary profiles. Ph.D. applicant?s interest in social robotics is required.

*Keywords* Human-Machine Interaction, Social Robotics, Deep Learning, Social Computing, Natural Language Processing, Speech Processing, Computer Vision, Multimodality

*Supervision* :  
Industrial: Marine Chamoux (Softbank robotics),
Academic: Chloé Clavel [Clavel],  Giovanna Varni [Varni] (Telecom-Paris)

*How to apply*
Applications should be sent as soon as possible (the first review of applications will be made in early September). The application should be formatted as **a single pdf file** and should include:
?    A complete and detailed curriculum vitae
?    A letter of motivation
?   The academic credentials and the transcript of grades
?    The contact of two referees

The pdf file should be sent to the three supervisors: mchamoux@softbankrobotics.com, chloe.clavel@telecom-paristech.fr, giovanna.varni@telecom-paristech.fr



*Description*
Social robotics, and more broadly human-agent interaction is a field of human-machine interaction for which the integration of social behaviors is expected to have great potential. 'Socio-emotional behaviors' (emotions, social stances) include thus the role and the reactions of the user towards the robot during an interaction. These behaviors could be expressed differently depending:
-on the user (age, emotional state, ...): some users may have a dominant behavior with the robot, considering it a tool to achieve a goal. Others are more cooperative with the robot, they can be more friendly with it. Still others try to trap or 'troll' the robot.
-on the interaction context  (users do not behave in the same way when interacting with a pepper selling toys, or with a pepper bank secretary). Besides, in each of these situations, the robot must be able to adapt its behavior, and to provide a coherent interaction between the user and the robot, avoiding confusion and frustration.

This Ph.D. will focus on multimodal modeling for the prediction of the user's socio-emotional behaviors during interactions with a robot and on building an engine that is robust to real-life scenarios and different contexts. In particular, the Ph.D. candidate will address the following points:
- the encoding of contextual multimodal representations relevant for the modeling of socio-emotional behavior. Thanks to the robot, we have access to a lot of information on context (market, robot intention, demographics, multi or mono user interaction, etc.) that could be combined to our multimodal representation.
- the development and evaluation of models that take advantage of the complementarity of modalities in order to monitor the evolution of the user's socio-emotional behaviors during the interaction (e. g. taking into account the inherent sequentially of the interaction structure)
The models will be based on sequential neural approaches (recurrent networks) that integrate attention models as a continuation of the work done in [Hemamou] and [BenYoussef19].

Selected references of the team:
[Hemamou] L. Hemamou, G. Felhi, V. Vandenbussche, J.-C. Martin, C. Clavel, HireNet: a Hierarchical Attention Model for the Automatic Analysis of Asynchronous Video Job Interviews.  in AAAI 2019
[Garcia] Alexandre Garcia, Chloé Clavel, Slim Essid, Florence d?Alche-Buc, Structured Output Learning with Abstention: Application to Accurate Opinion Prediction, ICML 2018
[Clavel&Callejas] Clavel, C.; Callejas, Z., Sentiment analysis: from opinion mining to human-agent interaction, Affective Computing, IEEE Transactions on, 7.1 (2016) 74-93.
[Langlet] C. Langlet and C. Clavel, Improving social relationships in face-to-face human-agent interactions: when the agent wants to know user?s likes and dislikes , in ACL 2015
[Maslowski]  Irina Maslowski, Delphine Lagarde, and Chloé Clavel.  In-the-wild chatbot corpus: from opinion analysis to interaction problem detection, ICNLSSP 2017.
[Ben-Youssef17]  Atef Ben-Youssef, Chloé Clavel, Slim Essid, Miriam Bilac, Marine Chamoux, and Angelica Lim.  Ue-hri: a new dataset for the study of user engagement in spontaneous human-robot interactions.  In  Proceedings of the 19th ACM International Conference on Multimodal Interaction, pages 464?472. ACM, 2017.
[Ben-Youssef19] Atef Ben Youssef; Chloe Clavel; Slim Essid Early Detection of User Engagement Breakdown in Spontaneous Human-Humanoid Interaction, IEEE Transactions on Affective Computing, 2019
[Varni] Varni G., Hupont, I., Clavel, C., Chetouani, M. Computational Study of Primitive Emotional Contagion in Dyadic Interactions. IEEE Transactions on Affective Computing, 2017.

[SB] https://www.softbankrobotics.com/emea/fr
[TP] https://www.telecom-paristech.fr/eng/  
[SocComp.] https://www.tsi.telecom-paristech.fr/recherche/themes-de-recherche/analyse-automatique-des-donnees-sociales-social-computing/
[SSA] http://www.tsi.telecom-paristech.fr/ssa/#
[Clavel] https://clavel.wp.imt.fr/publications/
[Varni] https://sites.google.com/site/gvarnisite/

Back

Top

Speech scientist at ETS Research :

https://etscareers.pereless.com/index.cfm?fuseaction=83080.viewjobdetail&CID=83080&JID=290092

Back

Top

Positions in Forensic Speech Science or Forensic Data Science:

- One Lecturer or Senior Lecturer

- Two Postdoctoral Researchers

 

Aston University, Birmingham, UK

 

Aston University has recently been awarded GBP 5.4 M from Research England?s Expanding Excellence in England (E3) Fund. The money is being used to expand the existing Centre for Forensic Linguistics into the substantially larger Aston Institute for Forensic Linguistics (AIFL). As part of the expansion, we are building a research team with expertise in forensic speech science and in forensic data science. In addition to conducting research in forensic speech science, members of the team will work on forensic inference and statistics more broadly, and on quantitative-measurement and statistical-model based approaches in other branches of forensic science. The latter potentially include but are not limited to: fingerprints, face, gait, ballistics, blood pattern analysis, and linguistics. The Forensic Speech Science Laboratory and the Centre for Forensic Data Science will be headed by Dr Geoffrey Stewart Morrison, and, in addition to the affiliation with AIFL, will be affiliated with the Computer Science Department in the School of Engineering and Applied Science.

 

We are looking to recruit the following positions:

 

Lecturer or Senior Lecturer in Forensic Speech Science or Forensic Data Science

Reference:    R190354

Salary:    Grade 9 £40,792 ? £48,677 or Grade 10 £50,132 ? £58,089

Contract Type:    Continuing 

Basis:    Full time

Closing Date:    23.59 hours BST on September 30, 2019

Interview Date:    To be confirmed

 

Two Postdoctoral Researchers in Forensic Speech Science or Forensic Data Science

Reference:    R190353

Salary:   Grade 8 £33,199 ? £39,609 or Grade 9 £40,792 ? £48,677

Contract Type:    Fixed term (3 years) 

Basis:    Full time or part time

Closing Date:    23.59 hours BST on September 30, 2019

Interview Date:    To be confirmed

 

The Lecturer or Senior Lecturer position will be a full-time permanent position and will include teaching and administrative responsibilities. The position is costed as a Grade 9 Lecturer, but an exceptionally well qualified and experienced successful applicant could potentially be appointed as a Grade 10 Senior Lecturer. Note: ?Lecturer? is equivalent to North American ?Assistant Professor?, ?Senior Lecturer? is equivalent to North American ?Associate Professor?, and ?Reader / Associate Professor? is an occasionally used additional rank between Senior Lecturer and Professor.

 

The Postdoctoral Researcher positions may be filled as full-time appointments (preferred) or via a combination of part-time appointments. The Postdoctoral Researcher positions will be fixed-term, but the plan is to build a team that will be successful in obtaining additional research funding that will sustain these positions.

 

All new team members must have a commitment to solving forensic problems. Previous experience working on forensic problems would be advantageous, but not essential. A background in forensic speech science, in other branches of forensic science, and/or in forensic inference and statistics would be advantageous, but not essential. At least one of the new team members must have a strong background in state-of-the-art automatic speaker recognition, with an ability to implement systems. Other useful backgrounds for members of the team would include biometrics, machine learning, natural language processing, and acoustic phonetics.

 

Candidates may apply for both the Lecturer / Senior Lecturer and the Research Associate positions. If positions are not filled after this round of recruitment, we will initiate another round of recruitment.

 

We also welcome enquiries from individuals who have obtained or are applying for their own postdoctoral fellowships, e.g., Marie Sklodowska-Curie Fellowships. For suitable candidates we would assist with the application process.

 

Potential candidates are encouraged to contact Dr Geoffrey Stewart Morrison to seek more information about these positions.

Tel:    +44 121 204 3901

e-mail:    g.s.morrison@aston.ac.uk

 

Dr Morrison will be attending Interspeech in September and would be happy to meet informally with potential applicants there.

 

Please visit our website http://www.aston.ac.uk/jobs for further information and to apply online.

 

Aston University is an equal opportunities employer and welcomes applications from all sections of the community.

Back

Top

There is a fully funded PhD position open at Idiap Research Institute on spiking neural
architectures for speech prosody.

The research will build on work done recently at Idiap on creating tools for
physiologically plausible modelling of speech. The current 'toolbox' contains rudimentary
muscle models and means to drive these using conventional (deep) neural networks. The
main focus of the work will involve use of spiking neural networks such as the 'integrate
and fire' type that is broadly representative of those found in biological systems.
Whilst we have focused so far on prosody (actually intonation), the application is open
ended; the focus is on the neural modelling. A key problem to be solved will be that of
training of the spiking networks, especially with the recurrence that is usual in such
networks. We hope to be able to train and use spiking networks as easily as conventional
backpropagation networks, and to shed light on current understanding of how biological
spiking networks learn (e.g., via spike timing-dependent plasticity).

For more information, and to apply, please follow this link:
 http://www.idiap.ch/education-and-jobs/job-10263

Idiap is located in Martigny in French speaking Switzerland, but functions in English and
hosts many nationalities.  PhD students are registered at EPFL. All positions offer quite
generous salaries.  Martigny has a distillery and a micro-brewery and is close to all
manner of skiing, hiking and mountain life.

There are other open positions on Idiap's main page
 https://www.idiap.ch/en/join-us/job-opportunities

Back

Top

Applications are invited for a three-year Early Stage Researcher PhD positions in the speech technology for pathological speech.

Back

Top

Post-doc position: Pattern mining for Neural Networks debugging: application to speech recognition

Back

Top

Cobalt Speech & Language (http://www.cobaltspeech.com/ ) is looking for a speech technologist/linguist to help find and create language resources for a project in French Canadian.

The project is short term (<2 months) part time (~5-8h  a week) which is ideal for a student to get an experience with speech industry.

The following skills are required:
  - Native French (does not have to be Canadian - though desirable)
  - Able to communicate in English
  - Basic understanding of speech technology and linguistics
  - Ability to run a python script.

For more information, please contact (Rasmus Dall) rasmus@cobaltspeech.com

Back

Top

PhD thesis proposal

Incremental sequence-to-sequence mapping for

speech generation using deep neural networks

September 4, 2019

1 Context and objectives

In recent years, deep neural networks have been widely used to address sequence-

to-sequence (S2S) learning. S2S models can solve many tasks where source

and target sequences have different lengths such as: automatic speech recog-

nition, machine translation, speech translation, text-to-speech synthesis, etc.

Recurrent, convolutional and transformer architectures, coupled with attention

models, have shown their ability to capture and model complex temporal de-

pendencies between a source and a target sequence of multidimensional discrete

and/or continuous data. Importantly, end-to-end training alleviates the need

to previously extract handcrafted features from the data by learning hierarchi-

cal representations directly from raw data (e.g. character string, video, speech

waveform, etc.).

The most common models are composed of an encoder that reads the full in-

put sequence (i.e. from its beginning to its end) before the decoder produces the

corresponding output sequence. This implies a latency equals to the length of

the input sequence. In particular, for a text-to-speech (TTS) system, the speech

waveform is usually synthesized from a complete text utterance (e.g. a sequence

of words with explicit begin/end-of-utterance markers). Such approach cannot

be used in a truly interactive scenario, in particular by a speech-handicapped

person to communicate orally'. Indeed, the interlocutor has to wait for the

complete utterance to be typed before being able to listen to the synthetic voice,

hence limiting the dynamics and naturalness of the interaction.

The goal of this project is to develop a general methodology for incremental

sequence-to-sequence mapping, with application to interactive speech technolo-

gies. It will require the development of end-to-end classication and regression

neural models able to deliver chunks of output data on-the-y, from only a par-

tial observation of input data. The goal is to learn an ecient policy that leads

 to an optimal trade-off between (variable) latency and accuracy of the decoding

process. Possible strategies to decode the output data as soon as possible in-

clude: (i) Predicting online he future' of the output sequence from he past

and present' of the input sequence, with an acceptable tolerance to possible er-

rors, or (2) learn automatically from the data an optimal waiting policy' that

prevents the model to output data when the uncertainty is too high. The devel-

oped methodology will be applied to address two speech processing problems:

(i) Incremental Text-to-Speech synthesis in which speech is synthesized while

the user is typing the text (possibly with a variable latency), and (ii) Incremen-

tal speech enhancement/inpainting in which portions of the speech signal are

unintelligible because of sudden noise or speech production disorders, and must

be replaced on-the-y with reconstructed portions.

2 Work plan

The proposed working plan is the following :

 Bibliographic work on S2S neural models, in the context of speech recogni-

tion, speech synthesis, and machine translation as well as their incremental

(low-latency) variations

 Investigating new architectures, losses, and training strategies toward in-

cremental S2S models.

 Implementing and evaluating the proposed techniques in the context of

end-to-end neural TTS systems (the baseline system may be a neural

TTS trained with past information/left-context only).

 Implementing and evaluating the proposed techniques in the context of

speech enhancement/inpainting, rst on simulated noisy speech and then

on pathological speech.

3 Requirements

We are looking for an outstanding and highly motivated PhD candidate to work

on this subject. Following requirements are mandatory:

 Engineering degree and/or a Master's degree in Computer Science, Signal

Processing or Applied Mathematics.

 Solid skills in Machine Learning. General knowledge in natural language

processing and/or speech processing.

 Excellent programming skills (mostly in Python and deep learning frame-

works).

 Good oral and written communication in English.

 Ability to work autonomously and in collaboration with supervisors and

other team members.

2

4 Work context

Grenoble Alpes Univ. o
ers an excellent research environment with ample com-

puting facilities, as well as remarkable surroundings to explore over the week-

ends. The PhD project will be funded by the Grenoble Artificial Intelligence

Institute (MIAI). The PhD candidate will work both at GIPSA-lab (CRISSP

team) and LIG-lab (GETALP team). The duration of the PhD is 3 years. The

salary is between 1770 and 2100 euros gross per month (depending on comple-

mentary activity or not).

5 How to apply?

Applications should include a detailed CV; a copy of their last diploma; at least

two references (people likely to be contacted); a cover letter of one page; a one-

page summary of the Master thesis; the two last transcripts of notes (Master or

engineering school). Applications should be sent to thomas.hueber@gipsa-lab.fr,

laurent.girin@gipsa-lab.fr and laurent.besacier@imag.fr. Applications will be

evaluated as they are received: the position is open until it is filled.

Back

Top

Postdoc proposal

Spontaneous Speech Recognition.

Application to Situated Corpora in French.

September 4, 2019

1 Postdoc Subject

The goal of the project is to advance the state-of-the-art in spontaneous auto-

matic speech recognition (ASR). Recent advances in ASR show excellent per-

formances on tasks such as read speech ASR (Librispeech), TV shows (MGB

challenge), but what about spontaneous communicative speech ?

This postdoc project would leverage existing transcribed corpora in French

(more than 300 hours) recorded in everyday communication (speech recordings

inside a family, in a shop, during an interview, etc.). One impact of the project

would be the automatization of transcription on very challenging data in order

to feed linguistic and phonetic studies at scale.

Research topics:

 End-to-end ASR models

 Spontaneous speech ASR

 Colloquial speech transcription

 Data augmentation for spontaneous and colloquial language modelling

 Transcribing situated corpora

2 Requirements

We are looking for an outstanding and highly motivated postdoc candidate to

work on this subject. Following requirements are mandatory:

 PhD degree in natural language processing or speech processing.

 Excellent programming skills (mostly in Python and deep learning frame-

works).

1

 Interest in pluri-disciplinary research (speech technology and speech sci-

ence)

 Good oral and written communication in English (French is a plus while

not mandatory)

 Ability to work autonomously and in collaboration with other team mem-

bers.

3 Work context

Grenoble Alpes Univ. o
ers an excellent research environment with ample com-

puting facilities, as well as remarkable surroundings to explore over the week-

ends. The postdoc project will be funded by the Grenoble Articial Intelligence

Institute (MIAI). The candidate will work both at LIG-lab (GETALP team)

and LIDILEM-lab. The duration of the postdoc is 18 months.

4 How to apply?

Applications should include a detailed CV; a copy of the last diploma; at least

two references (people likely to be contacted); a cover letter of one page; a

one-page summary of the PhD thesis. Applications should be sent to lau-

rent.besacier@imag.fr Applications will be evaluated as they are received: the

position is open until it is lled.

Back

Top

nceVOXCRIM 2019

MARDI 24 SEPTEMBRE 2019

de 9h30 à 17h00

Conférences et table ronde :

regards croisés sur la comparaison de voix

en criminalistique.

Inscriptions avant le 13 septembre

voxcrim@interieur.gouv.fr

04 72 86 85 22

Service Central de la Police

Technique et Scientifique

31 avenue Franklin Roosevelt

69130 ECULLY

Back

Top

The Social Computing Group at Idiap is seeking a creative and motivated postdoctoral
researcher to work on deep learning methods for behavioral analysis from video and audio
data. This is an opening for a researcher with experience in deep learning applied to
dynamic human behavior (from voice, body, or face), in the context of a project funded by
Innosuisse, the Swiss funding agency for promotion of innovation.

The position offers the opportunity to do exciting work on deep learning and social
behavior. The researcher will work with Prof. Daniel Gatica-Perez and his research group.
The candidates will have a PhD degree in computer science or engineering, with proven
experience in deep learning and a strong publication record.

Salaries are competitive and starting date is immediate. Interviews will start upon
reception of applications until the positions are filled.

Interested candidates are invited to submit a cover letter, a detailed CV, and the names
of three references through Idiap's online recruitment system:

https://www.idiap.ch/en/join-us/job-opportunities
Position: Postdoctoral researcher in deep learning for social behavior analysis

Interested candidates can also contact Prof. Daniel Gatica-Perez (gatica@idiap.ch).

About Idiap Research Institute

Idiap is an independent, not-for-profit, research institute recognized and funded by the
Swiss Federal Government, the State of Valais, and the City of Martigny. Idiap is an
equal opportunity employer, and offers competitive salaries and excellent working
conditions in a dynamic and multicultural environment.

Idiap is located in the town of Martigny in Valais, a scenic region in the south of
Switzerland, surrounded by the highest mountains of Europe, and offering exceptional
quality of life, exciting recreational activities, including hiking, climbing and skiing,
as well as varied cultural activities, all within close proximity to Lausanne and Geneva.
English is the official working language.

Back

Top

Postdoctoral Research Fellow/Senior Research Fellow

(speech and language technology, cognitive science)

Tampere University and Tampere University of Applied Sciences create

a unique environment for multidisciplinary, inspirational and highimpact

research and education. Our universities community has its

competitive edges in technology, health and society. www.tuni.fi/en

Speech and Cognition research group (SPECOG) is part of Computing Sciences Unit of Tampere University

within the Faculty of Information Technology and Communication Sciences. SPECOG focuses on

interdisciplinary research at the intersection of speech technology and cognitive sciences. We apply

advanced signal processing and machine learning methods to computational modeling of human language

learning and perception and study how human-like information processing principles can be applied in

autonomous next-generation artificial intelligence (AI) systems. The group also conducts research and

development in speech and language technology and in medical signal processing and machine learning.

SPECOG collaborates with several internationally leading research groups within and across disciplinary

boundaries, including joint research with computer scientists, psychologists, brain researchers, and linguists.

The group is also closely affiliated with audio and machine vision research groups of Tampere University.

More information on SPECOG: http://www.cs.tut.fi/sgn/specog/index.html

Job description

We are inviting applications for the position of a postdoctoral research fellow or senior research fellow in

the areas of speech and language technology and cognitive science. The work will be conducted as a

member of the SPECOG research group led by Asst. Prof. Okko Räsänen. We are looking for candidates who

are interested in human and/or machine language processing, and who are willing to contribute to our

highly cross-disciplinary research efforts in understanding language learning in humans and autonomous

computational systems. Our current focus is on machine learning algorithms for unsupervised language

learning from purely acoustic or audiovisual data (sometimes also known as zero-resource speech

processing). However, we also consider candidates with a strong independent research agenda in

complementary areas of speech and language technology.

In this position, the candidate is expected to:

1) carry out world-class research on a topic related to SPECOG focus areas

2) work in close collaboration with other members of the research group, and

3) help to advise undergraduate and/or PhD projects on the relevant topics (with flexibility according to

personal interests and career aspirations).

Requirements

The candidate should hold a doctoral degree (e.g., PhD or D.Sc. (Tech.)) in language technology, computer

science, electrical engineering, cognitive science, or other relevant area. Candidates who have already

completed their doctoral research work but have not yet received their doctoral certificate may also apply.

A successful candidate has strong expertise in signal processing and machine learning (e.g., deep learning),

ideally from the context of speech technology. Applicants with a background in natural language processing

(NLP) or cognitive science are also considered. Experience or interests in linguistics, neuroscience, or

statistics are considered as an advantage. Fluent programming (Python, Matlab, R, C++ or similar) and

English skills are required.

Potential candidates must be capable of carrying out independent research at the highest international

level. Competence must be demonstrated through several existing publications in internationally

recognized peer-reviewed journals and conferences.

We offer

The position will be filled for a fixed-term period of two years, starting as soon as possible (but not

extending the contract beyond the end of December 2021). A trial period of 6 months is applied to all new

employees. The exact starting date is negotiable.

We offer competitive academic salary, typically between 3500–4000 € for a starting postdoc depending on

the experience of the candidate, and 4000–4500 € for a senior research fellow with several years of existing

postdoctoral research experience in academia or industry. In addition, the position comes with extensive

benefits such as occupational healthcare, excellent sports facilities, flexible working hours, and several

restaurants and cafés on the campus with staff discounts. Traveling costs and daily allowances related to

presenting peer-reviewed work in major international conferences is also normally covered.

How to apply

Send the application through the online portal at https://tuni.rekrytointi.com/paikat/?o=A_A&jid=301

We will accept applications until the position has been filled, but no later than 30th of November 2019 at

23.59 (GMT+3). Note that we will start evaluating the applicants already on 1st of October 2019, and the

position may be filled as soon as a suitable candidate is found. We reserve the opportunity to recruit the

candidate through other channels or to decide to not to fill the position in case a suitable candidate is not

found during the process.

The application should contain the following documents (all in .pdf format):

- A free-form letter of motivation for the position in question (max. 1 page)

- Academic CV with contact information

- A list of publications

- A copy of doctoral degree certificate

- A letter or letters of recommendation (max. 3)

Please name all the documents as surname_CV.pdf, surname_list_of_publications.pdf … etc. Only the

applications sent through the university application portal and containing the requested attachments in the

instructed format will be considered in the recruitment process.

The most promising candidates will be interviewed in person or via Skype before the final decision.

For more information about the position, please contact Assistant Professor Okko Räsänen

(firstname.surname@tuni.fi; no umlauts) by email.

About the research environment

Finland is among the most stable, free and safe countries in the world, based on prominent ratings by

various agencies. It is also ranked as one of the top countries as far as social progress is concerned.

Tampere is counted among the major academic hubs in the Nordic countries and offers a dynamic living

environment. Tampere region is one of the most rapidly growing urban areas in Finland and home to a

vibrant knowledge-intensive entrepreneurial community. The city is an industrial powerhouse that enjoys a

rich cultural scene and a reputation as a centre of Finland’s information society. Despite its growth, living in

Tampere is highly affordable with two-room apartment rents starting from approx. 550 €. In addition, the

excellent public transport network enables quick, easy and cheap transportation around the city of

Tampere and university campuses.

Read more about Finland and Tampere:

• https://www.visitfinland.com/about-finland/

• https://finland.fi/

• http://julkaisut.valtioneuvosto.fi/bitstream/handle/10024/161193/MEAEguide_18_2018_T

ervetuloaSuomeen_Eng_PDFUA.pdf

• https://visittampere.fi/en/

Back

Top

Doctoral Researcher

(speech and language technology, cognitive science)

Tampere University and Tampere University of Applied Sciences create

a unique environment for multidisciplinary, inspirational and highimpact

research and education. Our universities community has its

competitive edges in technology, health and society. www.tuni.fi/en

Speech and Cognition research group (SPECOG) is part of Computing Sciences Unit of Tampere University

within the Faculty of Information Technology and Communication Sciences. SPECOG focuses on

interdisciplinary research at the intersection of speech technology and cognitive sciences. We apply

advanced signal processing and machine learning methods to computational modeling of human language

learning and perception and study how human-like information processing principles can be applied in

autonomous next-generation artificial intelligence (AI) systems. The group also conducts research and

development in speech and language technology and in medical signal processing and machine learning.

SPECOG collaborates with several internationally leading research groups within and across disciplinary

boundaries, including joint research with computer scientists, psychologists, brain researchers, and linguists.

The group is also closely affiliated with audio and machine vision research groups of Tampere University.

More information on SPECOG: http://www.cs.tut.fi/sgn/specog/index.html

Job description

We are inviting applications for the position of a doctoral researcher (doctoral student) in the areas of

speech and language technology and cognitive science. The work will be conducted as a member of the

SPECOG research group led by Asst. Prof. Okko Räsänen. We are looking for candidates who are interested

in human and/or machine language processing, and who are willing to contribute to our highly crossdisciplinary

research efforts in understanding language learning in humans and autonomous computational

systems. Our current focus is on machine learning algorithms for unsupervised language learning from

purely acoustic or audiovisual data (sometimes also known as zero-resource speech processing). However,

we also consider candidates with interest towards complementary areas of speech and language

technology.

In this position, the candidate is expected to:

1) carry out research on a mutually agreed topic

2) complete a doctoral degree, including mandatory course studies for a D.Sc. (tech.) degree

3) participate to doctoral program

4) be available for assisting tasks in teaching and research group activities (max. 15% of working time)

Requirements

The candidate should hold a master’s degree in language technology, computer science, electrical

engineering, mathematics, cognitive science, or other relevant technical area. Candidates who have already

completed their master’s studies but are graduating during 2019 may also apply. Exceptional master’s

students of Tampere University, who are close to graduation, can be also considered for the position. In

this case, the candidate is first employed as a Research Assistant to carry out a master’s thesis project (6

months) on the topic and, upon a successful thesis project, with the possibility to continue to doctoral

studies.

A successful candidate has experience from signal processing and/or machine learning (e.g., deep learning),

ideally from the context of speech technology. Applicants with a background in natural language processing

(NLP) or cognitive science are also considered. Experience or interests in linguistics, neuroscience, or

statistics are considered as an advantage but not required. Good command of programming (Python,

Matlab, R, C++ or similar) and English skills are required.

Potential candidates must be capable of carrying out independent research work but are also good team

players. Previous experience from research such as research internships or other research projects are

considered as a significant advantage.

We offer

The position will be filled for a fixed-term period of two years with a view for extension. A trial period of 6

months is applied to all new employees. The position start in January 2020 or as soon as possible with a

negotiable exact starting date. Target completion time for doctoral studies is 4 years.

We offer a starting salary of 2300 € for a starting doctoral researcher with later increases based on

demonstrated progress through scientific publications and acquired study credits. In addition, the position

comes with extensive benefits such as occupational healthcare, excellent sports facilities, flexible working

hours, and several restaurants and cafés on the campus with staff discounts. Traveling costs and daily

allowances related to presenting peer-reviewed work in major international conferences is also normally

covered.

How to apply

Send your application through the online portal at https://tuni.rekrytointi.com/paikat/?o=A_A&jid=299

We will accept applications until 15th of November 2019 at 23.59 (GMT+3). We reserve the opportunity to

recruit the candidate through other channels or to decide to not to fill the position in case a suitable

candidate is not found during the process.

The application should contain the following documents (all in .pdf format):

- A free-form letter of motivation for the position in question (max. 1 page)

- Complete CV with contact information and a list of publications (if any)

- A copy of master’s degree certificate

- English language certificate of proficiency (for non-native and non-Finnish applicants)

Please name all the documents as surname_CV.pdf, surname_list_of_publications.pdf … etc. Only the

applications sent through the university application portal and containing the requested attachments in the

instructed format will be considered in the recruitment process.

The most promising candidates will be interviewed in person or via Skype before the final decision.

For more information about the position, please contact Assistant Professor Okko Räsänen

(firstname.surname@tuni.fi; no umlauts) by email.

About the research environment

Finland is among the most stable, free and safe countries in the world, based on prominent ratings by

various agencies. It is also ranked as one of the top countries as far as social progress is concerned.

Tampere is counted among the major academic hubs in the Nordic countries and offers a dynamic living

environment. Tampere region is one of the most rapidly growing urban areas in Finland and home to a

vibrant knowledge-intensive entrepreneurial community. The city is an industrial powerhouse that enjoys a

rich cultural scene and a reputation as a centre of Finland’s information society. Despite its growth, living in

Tampere is highly affordable with private market two-room apartment rents starting from approx. 550 €. In

addition, the excellent public transport network enables quick, easy and cheap transportation around the

city of Tampere and university campuses.

Read more about Finland and Tampere:

• https://www.visitfinland.com/about-finland/

• https://finland.fi/

• http://julkaisut.valtioneuvosto.fi/bitstream/handle/10024/161193/MEAEguide_18_2018_T

ervetuloaSuomeen_Eng_PDFUA.pdf

• https://visittampere.fi/en/

Back

Top

Projet READYNOV : AUDIOCAP

Audition et handicap dans le bruit : vers la restauration de l’intelligibilité de la parole

Type d’emploi POSTDOC

Cadre de la recherche

Restauration d’une intelligibilité dans le bruit pour les personnes

âgées via des prothèses auditives.

Mots-clés Parole, bruit, intelligibilité

Missions

Prédiction de l’intelligibilité de la parole dans le bruit :

- Prise en main d’un système de Reconnaissance Automatique de

la Parole en Français,

- Modélisation acoustique dans le bruit.

Mise en place d’un outil de séparation de la parole et du bruit, fondée sur

l’application de filtres temps-fréquences. Celui-ci sera « réglé » dans un

but de favoriser l’intelligibilité de la parole.

Compétences

Développement logiciel

Traitement du signal

Apprentissage machine (« deep learning »)

Lieu IRIT – 118, route de Narbonne – 31062 TOULOUSE

Date et durée de la mission De 12 à 18 mois à partir du 1er octobre 2019

Salaire Entre 1900 et 2400 € net par mois, suivant l’expérience

Documents à fournir

- CV détaillé

- Lettre de motivation

- Résumé d'une page de la thèse de doctorat

Contact Julien PINQUIER, pinquier@irit.fr

Back

Top

ZAION est une société innovante en pleine expansion spécialisée dans la technologie des robots conversationnels : callbot et chatbot intégrant de l’Intelligence Artificielle.

ZAION a développé une solution qui s’appuie sur une expérience de plus de 20 ans de la Relation Client. Cette solution en rupture technologique reçoit un accueil très favorable au niveau international et nous comptons déjà 18 clients actifs (GENERALI, MNH, APRIL, CROUS, EUROP ASSISTANCE, PRO BTP …).

Nous sommes actuellement parmi les seuls au monde à proposer une offre de ce type entièrement tournée vers la performance. Nous rejoindre, c’est prendre part à une aventure passionnante au sein d’une équipe ambitieuse afin de devenir la référence sur le marché des robots conversationnels.

Nous rejoindre, c’est prendre part à une aventure passionnante et innovante afin de devenir la référence sur le marché des robots conversationnels. Dans le cadre de son développement ZAION recrute son Data Scientist /Machine Learning appliqué à l’Audio H/F. Au sein de l’équipe R&D, votre rôle est stratégique dans le développement et l’expansion de la société. Vous développerez, une solution qui permet de détecter les émotions dans les conversations. Nous souhaitons augmenter les fonctionnalités cognitives de nos callbots afin qu’ils puissent détecter les émotions de leurs interlocuteurs (joie, stress, colère, tristesse…) et donc adapter leurs réponses en conséquence.

Vos missions principales :

- Vous participez à la création du pôle R&D de ZAION et piloterez à votre arrivée votre premier projet de reconnaissance d’émotion dans la voix.

- Construisez, adaptez et faites évoluer nos services de détection d’émotion dans la voix

- Analysez de bases de données conséquentes de conversations pour en extraire les conversations émotionnellement pertinentes

- Construisez une base de données de conversations labelisées avec des étiquettes émotionnelles

- Formez et évaluez des modèles d'apprentissage automatique pour la classification d’émotion

- Déployez vos modèles en production

- Améliorez en continue le système de détection des émotions dans la voix

Qualifications requises et expérience antérieure :

-Vous avez une expérience de 2 ans minimum comme Data Scientist/Machine Learning appliqué à l’Audio

- Diplômé d’une école d’Ingénieur ou Master en informatique ou un doctorat en informatique mathématiques avec des compétences solides en traitements de signal (audio de préférence)

- Solide formation théorique en apprentissage machine et dans les domaines mathématiques pertinents (clustering, classification, factorisation matricielle, inférence bayésienne, deep learning...)

- La mise à disposition de modèles d'apprentissage machine dans un environnement de production serait un plus

- Vous maîtrisez un ou plusieurs des langages suivants : Python, Frameworks de machine Learning/Deep Learning (Pytorch, TensorFlow,Sci-kit learn, Keras) et Javascript

- Vous maîtrisez les techniques du traitement du signal audio

- Une expérience confirmée dans la labélisation de grande BDD (audio de préférence) est indispensable ;

- Votre personnalité : Leader, autonome, passionné par votre métier, vous savez animer une équipe en mode projet

- Vous parlez anglais couramment

Merci d’envoyer votre candidature à : alegentil@zaion.ai

Back

Top

Post-doctoral position (1 year): Perception for interaction and social navigation

Research Engineer (1 year): Social Human-Robot Interactions

Laboratory: LITIS, INSA Rouen Normandy, France

Project: INCA (Natural Interactions with Artificial Companions)

Summary:

The emergence of interactive robots and connected objects has lead to the appearance of symbiotic systems made up of human users, virtual agents and robots in social interactions. However, two major scientific difficulties are unsolved yet: on the one hand, the recognition of human activity remains inaccurate, both at the operational level (location, mapping and identification of objects and users) and cognitive (recognition and tracking of users? intentions) and, on the other hand, interaction involves different modalities that must be adapted according to the context, the user and the situation. The INCA project aims at developing artificial companions (interactive robots and virtual agents) with a particular focus on social interactions. Our goal is to develop new models and algorithms for intelligent companions capable of (1) perceiving and representing an environment (real, virtual or mixed) consisting of objects, robots and users; (2) interacting with users in a natural way to assess their needs, preferences, and engagement; (3) learning models of user behavior and (4) generating semantically adequate and socially appropriate responses.

Post-doctoral position in perception for interaction and social navigation (1 position)

The candidate will work to ensure that a robot can recognize the physical content of the scene surrounding him, recognize himself, static and dynamic objects (users and other robots) and finally predict the movement of dynamic elements. The integration of data from different sensors should allow the mapping of an unknown environment and estimate the position of the robot. First, VSLAM techniques (Visual Simultaneous Localization And Mapping) (Saputra 2018) will be used to map the scene. The regions (or points) of interest detected could
then be used to detect obstacles. In order to distinguish between static and dynamic objects, methods of separating the background from the foregound of the scene (Kajo et al, 2018) will be used. Finally, some recent techniques of the Flownet 2.0 type (Eddy et al, 2017), for the prediction of the motion on a video sequence should make it possible to predict the next movement of an object dynamic object and the to apprehend its behavior.

Profile: the candidate must have strong skills in mobile robotics and navigation techniques (VSLAM, OrbSlam, Optical Flow, stereovision...) and a high programming capacities under ROS or any other programming language compatible with robotics. Machine learning and Deep learning skills will be highly appreciated.

Research Engineer in Social Human-Robot Interactions (1 position)

The hired research engineer will work closely with the INCA research staff (permanent, PhD and post-doctoral members) and other project partners. This will mainly involve administering the project's Pepper robots, developing the necessary tools, integrating the algorithms developed with the AgentSlang platform (https://agentslang.github.io/) and join the team created to participate in the Robocup 2020 in Bordeaux, @Home league.

Profile: Computer Sciences / Robotics Engineer

• Good level in programming (ROS, Python, possibly Java)

• Strong knowledge in robotics

• Experiences in some of the following areas would be a plus (non-exhaustive list): machine learning, human-machine social interactions, scene perception, spatio-temporal and semantic representation, natural language dialogue.

Duration and remuneration: 1 year, 2480euros/month (gross salary)

Application should be sent to: alexandre.pauchet@insa-rouen.fr

• Curriculum vitae

• Cover letter

• Recommendation letters

• Recently graduated students: transcripts

Back

Top

un poste d'ATER en Informatique est disponible à la faculté des lettres de Sorbonne
Université. Le lien pour postuler est http://lettres.sorbonne-universite.fr/ater

Back

Top

Postdoc proposal

Spontaneous Speech Recognition.

Application to Situated Corpora in French.

September 4, 2019

1 Postdoc Subject

The goal of the project is to advance the state-of-the-art in spontaneous auto-

matic speech recognition (ASR). Recent advances in ASR show excellent per-

formances on tasks such as read speech ASR (Librispeech), TV shows (MGB

challenge), but what about spontaneous communicative speech ?

This postdoc project would leverage existing transcribed corpora in French

(more than 300 hours) recorded in everyday communication (speech recordings

inside a family, in a shop, during an interview, etc.). One impact of the project

would be the automatization of transcription on very challenging data in order

to feed linguistic and phonetic studies at scale.

Research topics:

 End-to-end ASR models

 Spontaneous speech ASR

 Colloquial speech transcription

 Data augmentation for spontaneous and colloquial language modelling

 Transcribing situated corpora

2 Requirements

We are looking for an outstanding and highly motivated postdoc candidate to

work on this subject. Following requirements are mandatory:

 PhD degree in natural language processing or speech processing.

 Excellent programming skills (mostly in Python and deep learning frame-

works).

1

 Interest in pluri-disciplinary research (speech technology and speech sci-

ence)

 Good oral and written communication in English (French is a plus while

not mandatory)

 Ability to work autonomously and in collaboration with other team mem-

bers.

3 Work context

Grenoble Alpes Univ. o
ers an excellent research environment with ample com-

puting facilities, as well as remarkable surroundings to explore over the week-

ends. The postdoc project will be funded by the Grenoble Articial Intelligence

Institute (MIAI). The candidate will work both at LIG-lab (GETALP team)

and LIDILEM-lab. The duration of the postdoc is 18 months.

4 How to apply?

Applications should include a detailed CV; a copy of the last diploma; at least

two references (people likely to be contacted); a cover letter of one page; a

one-page summary of the PhD thesis. Applications should be sent to lau-

rent.besacier@imag.fr Applications will be evaluated as they are received: the

position is open until it is lled.

Back

Top

PhD thesis proposal

Incremental sequence-to-sequence mapping for

speech generation using deep neural networks

September 4, 2019

1 Context and objectives

In recent years, deep neural networks have been widely used to address sequence-

to-sequence (S2S) learning. S2S models can solve many tasks where source

and target sequences have di
erent lengths such as: automatic speech recog-

nition, machine translation, speech translation, text-to-speech synthesis, etc.

Recurrent, convolutional and transformer architectures, coupled with attention

models, have shown their ability to capture and model complex temporal de-

pendencies between a source and a target sequence of multidimensional discrete

and/or continuous data. Importantly, end-to-end training alleviates the need

to previously extract handcrafted features from the data by learning hierarchi-

cal representations directly from raw data (e.g. character string, video, speech

waveform, etc.).

The most common models are composed of an encoder that reads the full in-

put sequence (i.e. from its beginning to its end) before the decoder produces the

corresponding output sequence. This implies a latency equals to the length of

the input sequence. In particular, for a text-to-speech (TTS) system, the speech

waveform is usually synthesized from a complete text utterance (e.g. a sequence

of words with explicit begin/end-of-utterance markers). Such approach cannot

be used in a truly interactive scenario, in particular by a speech-handicapped

person to communicate orally'. Indeed, the interlocutor has to wait for the

complete utterance to be typed before being able to listen to the synthetic voice,

hence limiting the dynamics and naturalness of the interaction.

The goal of this project is to develop a general methodology for incremental

sequence-to-sequence mapping, with application to interactive speech technolo-

gies. It will require the development of end-to-end classication and regression

neural models able to deliver chunks of output data on-the-y, from only a par-

tial observation of input data. The goal is to learn an ecient policy that leads

to an optimal trade-o
between (variable) latency and accuracy of the decoding

process. Possible strategies to decode the output data as soon as possible in-

clude: (i) Predicting online he future' of the output sequence from he past

1

and present' of the input sequence, with an acceptable tolerance to possible er-

rors, or (2) learn automatically from the data an optimal waiting policy' that

prevents the model to output data when the uncertainty is too high. The devel-

oped methodology will be applied to address two speech processing problems:

(i) Incremental Text-to-Speech synthesis in which speech is synthesized while

the user is typing the text (possibly with a variable latency), and (ii) Incremen-

tal speech enhancement/inpainting in which portions of the speech signal are

unintelligible because of sudden noise or speech production disorders, and must

be replaced on-the-y with reconstructed portions.

2 Work plan

The proposed working plan is the following :

 Bibliographic work on S2S neural models, in the context of speech recogni-

tion, speech synthesis, and machine translation as well as their incremental

(low-latency) variations

 Investigating new architectures, losses, and training strategies toward in-

cremental S2S models.

 Implementing and evaluating the proposed techniques in the context of

end-to-end neural TTS systems (the baseline system may be a neural

TTS trained with past information/left-context only).

 Implementing and evaluating the proposed techniques in the context of

speech enhancement/inpainting, rst on simulated noisy speech and then

on pathological speech.

3 Requirements

We are looking for an outstanding and highly motivated PhD candidate to work

on this subject. Following requirements are mandatory:

 Engineering degree and/or a Master's degree in Computer Science, Signal

Processing or Applied Mathematics.

 Solid skills in Machine Learning. General knowledge in natural language

processing and/or speech processing.

 Excellent programming skills (mostly in Python and deep learning frame-

works).

 Good oral and written communication in English.

 Ability to work autonomously and in collaboration with supervisors and

other team members.

2

4 Work context

Grenoble Alpes Univ. o
ers an excellent research environment with ample com-

puting facilities, as well as remarkable surroundings to explore over the week-

ends. The PhD project will be funded by the Grenoble Articial Intelligence

Institute (MIAI). The PhD candidate will work both at GIPSA-lab (CRISSP

team) and LIG-lab (GETALP team). The duration of the PhD is 3 years. The

salary is between 1770 and 2100 euros gross per month (depending on comple-

mentary activity or not).

5 How to apply?

Applications should include a detailed CV; a copy of their last diploma; at least

two references (people likely to be contacted); a cover letter of one page; a one-

page summary of the Master thesis; the two last transcripts of notes (Master or

engineering school). Applications should be sent to thomas.hueber@gipsa-lab.fr,

laurent.girin@gipsa-lab.fr and laurent.besacier@imag.fr. Applications will be

evaluated as they are received: the position is open until it is lled.

Back

Top

offre de post-doctorat au sein de l’activité « Evaluation des systèmes d’intelligence artificielle » du LNE :

https://www.lne.fr/fr/offre-emploi/post-doc-evaluation-systemes-evolutifs-locuteur-traduction

Le candidat retenu intégrera une équipe en forte croissance spécialisée en évaluation des systèmes d’IA, ainsi qu’un projet européen ambitieux portant sur les systèmes de traitement de la langue évolutifs (en traduction et en diarisation). La caractérisation des performances des systèmes intelligents capables de s’auto-améliorer au fur et à mesure de leur utilisation, par eux-mêmes et par interaction avec l’humain utilisateur, représente un véritable défi que ce post-doctorat propose de relever.

Back

Top

   An open full-time Post-Doc employment

    position for 30 months in the context of our research project DyNaVoiceR

    which is supported by FCT (the Portuguese Foundation for Science and

    Technology).

    The official announcement can be found:

    English version:

http://www.eracareers.pt/opportunities/index.aspx?task=showAnuncioOportunities&jobId=118038&idc=1e

    Portuguese version:

http://www.eracareers.pt/opportunities/index.aspx?task=showAnuncioOportunities&jobId=118038&lang=pt&idc=1e

    The salary level is quite attractive: 2.128,34 Euros per month (14

    salaries per year)

Aapplication deadline is October 15.

Back

Top

Ingénieur de recherche en

Traitement Automatique du Langage – F/H

Poste en CDI

Localisation : Laboeratoire national de metrologie et d'essais,Trappes

Traitement Automatique du Langage – F/H

Référence : ML/ITAL/DEC

Leader dans l’univers de la mesure et des références, jouissant d’une forte notoriété en France et à l’international, le LNE soutient l’innovation industrielle et se positionne comme un acteur important pour une économie plus compétitive et une société plus sûre.

Au carrefour de la science et de l’industrie depuis sa création en 1901, le LNE offre son expertise à l’ensemble des acteurs économiques impliqués dans la qualité et la sécurité des produits.

Pilote de la métrologie française, notre recherche est au coeur de cette mission de service public et l’une des clés du succès des entreprises.

Nous nous attachons à répondre au besoin industriel et académique de mesures toujours plus justes, dans des conditions de plus en plus extrêmes ou sur les concepts les plus émergents tels que les véhicules autonomes, les nanotechnologies ou la fabrication additive.

Missions :

Vous intégrerez une équipe de six ingénieurs-docteurs régulièrement accompagnés de post-doctorants, doctorants et stagiaires, spécialisée dans l’évaluation et la qualification des systèmes d’intelligence artificielle. Cette équipe est historiquement reconnue pour son expertise dans l’évaluation des systèmes de traitement automatique du langage naturel et le poste proposé doit contribuer à renforcer cette expertise dans un contexte de forte dynamique de croissance.

Depuis quelques années, l’équipe s’est diversifiée en termes de domaines d’application de son expertise d’évaluation des intelligences en traitant de sujets tels que les dispositifs médicaux, les robots industriels collaboratifs, les véhicules autonomes, etc. L’équipe capitalise sur les savoir-faire à la fois divers et ciblés de ses experts (TAL, imagerie, robotique, etc.) afin d’apporter conjointement une solution satisfaisante à la question de l’évaluation et de la certification des systèmes intelligents, condition impérative de leur acceptabilité et faisant l’objet aujourd’hui d’une attention prioritaire des pouvoirs publics.

C’est dans le cadre de la mise en place progressive d’un centre d’évaluation des systèmes intelligents à vocation nationale et internationale qu’elle cherche à accueillir les meilleurs profils de chaque spécialité de l’IA. Les missions principales de ce futur centre sont le développement de nouveaux protocoles d’évaluation, la qualification et la certification de systèmes intelligents, l’organisation de challenges (campagnes de benchmarking), la mise à disposition de ressources expérimentales, le développement et l’organisation du secteur d’activité et la définition de principes, politiques, doctrines et normes à cet effet.

En tant qu’ingénieur-docteur de recherche en TAL, votre champ d’intervention prioritaire sera le traitement de la langue (texte et parole). Vous pourrez également être amené.e à intervenir dans d’autres domaines du traitement de l’information (par exemple sur le traitement de l’image dont la reconnaissance optique de caractères), puis au-delà en fonction des priorités et de vos propres compétences et affinités.

Le poste est évolutif sur le moyen et le long termes en ce qu’il vise à la formation d’experts techniques de stature au moins nationale et ayant vocation à mener eux-mêmes la politique de croissance et de tutelle de leur spécialité, sous réserve du cadre réglementaire et d’orientations générales du LNE ou de ses donneurs d’ordres.

Dans un premier temps, vous couvrirez les missions suivantes :

- Contribution à la R&D et aux actions structurantes (60%) :

 Inventaire technique et commercial du besoin et de l’offre, priorisation des marchés et champs techniques à investir

 Identification et définition des grandeurs à mesurer, des métriques afférentes, des protocoles d’évaluation et des moyens d’essais nécessaires

 Structuration des données de la discipline (TAL) au sein de référentiels et selon des nomenclatures à bâtir

 Programmation et conduite d’essais à des fins expérimentales, de recherche itérative et d’étalonnage

 Constitution et animation d’un réseau de chercheurs des secteurs public et privé, national et étranger, en appui aux présentes missions

 Contribution au montage et à l’exécution de projets de recherche nationaux et européens et de coopérations internationales

 Participation aux travaux de planification du LNE : investissements, RH, budgets annuels, perspectives pluriannuelles

 Publication et présentation des résultats scientifiques

 Encadrement éventuel de doctorants, post-doctorants, stagiaires

- Contribution aux prestations commerciales en TAL (40%) :

 Ingénierie linguistique générale (manipulation des données, analyse statistique, etc.)

 Prise en charge du besoin client et reformulation dans le cadre d’une offre technique et commerciale

 Organisation des tâches pour la réalisation de la prestation, estimation des ressources nécessaires, négociation

 Réalisation de ces tâches en coordination avec l’équipe

 Production/rédaction des livrables

 Présentation des résultats au client

Profil :

Vous êtes titulaire soit d’un doctorat, soit d’un diplôme d’ingénieur avec un minimum de trois ans d’expérience professionnelle, en informatique ou en sciences du langage, avec une spécialisation en traitement automatique de la langue (TAL) et plus généralement en intelligence artificielle. Les expériences professionnelles ou académiques passées en développement et/ou test logiciel, en analyse statistique, ainsi qu’en traitement de la parole ou de l’image seront particulièrement appréciées.

Vous disposez également d’un bon niveau d’anglais et de programmation (C++ et/ou Python), ainsi que d’une expérience en utilisation de Linux.

Dans le cadre de votre prise de poste, vous pourriez être amené.e à suivre des formations complémentaires (par exemple en intelligence artificielle et en cybersécurité).

Vous saurez être à l’initiative, en disposant d’une large autonomie et d’un potentiel de créativité vous permettant d’occuper pleinement votre espace de responsabilité dans un objectif d’excellence. Vous êtes capable de défendre un leadership de par la qualité et la clarté de vos argumentaires.

Déplacements fréquents en région parisienne (une fois par semaine), en province (une à deux fois par mois) et occasionnels dans le monde (une fois par trimestre) dans le cadre de prestations, réunions ou conférences.

Back

Top

Postdoctoral Researcher in Multilingual Speech Processing

CONTEXT The Expression research team focuses on expressiveness in human-centered data. In this context, the team has a strong activity in the field of speech processing, especially text-to-speech (TTS). This activity is denoted by regular publications in top international conferences and journals, exposing contributions in topics like machine learning (including deep learning), natural language processing, and speech processing. Team Expression takes part in multiple collaborative projects. Among those, the current position will take part in a large European H2020 project focusing on the social integration of migrants in Europe. Team’s website: https://www-expression.irisa.fr/

PROFILE Main tasks: 1. Design multilingual TTS models (acoustic models, grapheme-to-phoneme, prosody, text normalization...) 2. Take part in porting the team’s TTS system for embedded environments 3. Develop spoken language skill assessment methods

Secondary tasks: 1. Collect speech data 2. Define use cases with the project partners

Environment: The successful candidate will integrate a team of other researchers and engineers working on the same topics.