![]() |
ISCApad #271 |
Monday, January 11, 2021 by Chris Wellekens |
6-1 | (2020-07-09) Speech Research Scientist at ETS R&D Speech Research Scientist at ETS R&D:
https://etscareers.pereless.com/index.cfm?fuseaction=83080.viewjobdetail&CID=83080&JID=302819
| |||||
6-2 | (2020-07-16) Early Stage Researcher / PhD student in an EU Marie Sklodowska-Curie Action (H2020-MSCA-ITN), Romania We are hiring an Early Stage Researcher / PhD student in an EU Marie Sklodowska-Curie Action (H2020-MSCA-ITN) on the topic 'Designing and Engineering Multimodal Feedback to Augment the User Experience of Touch Input' under the supervision of Prof. Radu-Daniel Vatavu (http://www.eed.usv.ro/~vatavu)
| |||||
6-3 | (2020-07-20) Technical project manager, INRIA, Nancy, France Inria is seeking a Technical Project Manager for a European (H2020 ICT) collaborative
| |||||
6-4 | (2020-09-10) Experts in recognition and synthesis at Reykjavík University's Language and Voice Lab, IcelandReykjavík University's Language and Voice Lab (https://lvl.ru.is) is looking
for experts in speech recognition and in speech synthesis. At the LVL you
will be joining a research team working on exciting developments in language
technology as a part of the Icelandic Language Technology Programme
(https://arxiv.org/pdf/2003.09244.pdf).
Job Duties:
. Conduct independent research in the fields of speech processing,
machine learning, speech recognition/synthesis and human-computer interaction.
. Work with a team of other experts in carrying out the Speech
Recognition/Synthesis part of the Icelandic Language Technology Programme.
. Publish and disseminate research findings in journals and present at
conferences. . Actively take part in scientific and industrial cooperation projects.
. Assist in supervising Bachelor's/Master's students. Skills:
. MSc/PhD degree in engineering, computer science, statistics, mathematics
or similar
. Good programming skills (e.g. C++ and Python) and knowledge of Linux
(necessary).
. Good knowledge of a deep learning library such as PyTorch or TensorFlow
(necessary).
. Good knowledge of KALDI (preferable)
. Background in language technology (preferable).
. Good skills in writing and understanding shell scripts (preferable).
All applications must be accompanied by a good CV with information about
previous jobs, education, references etc. It is also optional to attach a cover
letter where the applicant can justify the reasons for being the right person
for the job. Here is the link to apply: https://jobs.50skills.com/ru/is/5484
Applications deadline is October 4th 2020.
Applications are only accepted through RU's recruitment system.
All inquiries and applications will be treated as confidential.
Further information about the job is provided by Jón Guðnason
Associate Professor, jg@ru.is, and Ester Gústavsdóttir, Director of Human
Resources, esterg@ru.is.
The role of Reykjavik University is to create and disseminate knowledge to
enhance the competitiveness and quality of life for individuals and society,
guided by good ethics, sustainability and responsibility.
Education and research at RU are based on strong ties with industry and society.
We emphasize interdisciplinary collaboration, international relations and
entrepreneurship.
| |||||
6-5 | (2020-09-17) Proposition de contrat doctoral, Sorbonne University (Jussieu), Paris, France Proposition de contrat doctoral Titre : Rythme de la parole et gestes manuels en synthèse performative Résumé du sujet : Le but de cette thèse est de développer un cadre théorique et des expérimentations quant à l'utilisation du geste manuel pour le contrôle prosodique via des interfaces humain-machine, en synthèse performative. La synthèse vocale performative est un nouveau paradigme de recherche en interaction humain-machine, dans lequel une voix de synthèse est jouée comme un instrument en temps-réel à l’aide des membres (mains, pieds). Le contrôle du rythme de parole par les mains est un problème qui implique des unités rythmiques, des points de contrôle rythmique, les centres perceptifs des syllabes et des gestes de tapotement (tapping), voire des partitions gestuelles inspirées des phonologies autosegmentales ou articulatoires. Les unités rythmiques varient en fonction de la phonologie de la langue étudiée, ici le français, l’anglais et le chinois mandarin. Les enjeux de la thèse portent donc sur la modélisation des schémas de perception-action impliqués dans le contrôle rythmique, la modélisation des unités temporelles, la réalisation et l’évaluation d’un système de contrôle du rythme. Les applications visées sont : 1. l’apprentissage du contrôle naturel des contours intonatifs à l'aide de la chironomie pour l'acquisition de langues étrangères (anglais, français, mandarin) ; 2. l’apprentissage du contrôle chironomique des contours d'intonation de la langue maternelle, pour la suppléance vocale (larynx artificiel). Contexte : La voix n’est pas un 'instrument' de musique, au sens d’un artefact mis en vibration par les membres ou par le souffle. Les organes vocaux sont internes, en grande partie invisibles, et contrôlés de façon complexe par plusieurs ensembles musculaires (respiration, phonation, articulation). Le contrôle vocal est donc, par nature intéroceptif, alors qu’il est davantage kinesthésique et extéroceptif pour les instruments de musique. L’avènement de la synthèse numérique permet pour la première le rendu d’un son indéniablement vocal par un dispositif instrumental externe, mis à distance de l’appareil vocal. Les 'instruments vocaux' sont 'manoeuvrés' par les mains, les pieds, à l’aide de capteurs ou d’interfaces humainmachine. Cette mise à distance pose la question du contrôle vocal dans des termes tout à fait différents de ceux du contrôle d’un instrument acoustique ou de la voix elle même. Les instruments vocaux permettent actuellement un contrôle musical de la phonation : intonation, séquencement rythmique, qualité de voix, pour la voix chantée. Le contrôle très précis de l’articulation et du rythme en parole est encore problématique. Le propos de cette thèse est de traiter la question du contrôle gestuel du rythme prosodique et du séquencement articulatoire. Objectifs et résultats attendus : Cette thèse s’inscrit dans la ligne de recherche sur les instruments vocaux. Un instrument vocal est un synthétiseur en vocal temps réel à contrôle gestuel. La synthèse est réalisée par un programme pour produire les échantillons. Le contrôle gestuel utilise des interfaces pour capter les gestes. Les mouvements des articulateurs étant très rapides, il est difficile de les contrôler de façon directe par les gestes manuels et des méthodologies basées sur la représentation phonologique du rythme prosodique doivent être mise en place. Le rythme est réalisé par des gestes des membres, mains ou pieds, en place des gestes articulatoire qui correspondent aux syllabes. Les circuits de perception-action ne sont plus les mêmes, ni les vélocités des organes mis en mouvement. Le contrôle du rythme prosodique en synthèse performative est donc un problème qui implique la définition d’unités rythmiques, de points de contrôle rythmique, de centres perceptifs des syllabes, de gestes de tapotement (tapping), voire de partitions gestuelles inspirées des phonologies autosegmentale ou articulatoire. Des points de contrôles rythmiques doivent enrichir le signal vocal pour permettre d’en manipuler le déroulement temporel. Ces points doivent avoir du sens du point de vue de la phonologie de la langue jouée, et de sa phonotactique. La perception du flux syllabique, avec ses centres perceptifs, est donc impliquée. Les gestes de contrôle, par appuis ou tapotage, impliquent des processus moteurs, à la fois analogues et différents de ceux des articulateurs. Les unités rythmiques varient en fonction de la phonologie des langues étudiées, ici le français, l’anglais et le chinois mandarin. Les enjeux de la thèse portent donc sur la modélisation des schémas de perception-action impliqués dans le contrôle rythmique, la modélisation des unités temporelles, la réalisation et l’évaluation d’un système de contrôle du rythme. Les résultats attendus sont à la fois théoriques et pratiques : • L’expérimentation perceptive permettra de mettre en relation les différentes unités temporelles; • les théories phonologiques sur l’organisation du geste phonatoire seront mises à l’épreuve avec un nouveau paradigme expérimental; • un nouveau synthétiseur sera réalisé; • un ensemble de méthodes pour le contrôle gestuel de la synthèse, de nouveaux gestes et des interfaces adaptées seront développés et testés dans les tâches applicative visées, soit l’apprentissage du contrôle naturel des contours intonatifs à l'aide de la chironomie pour l'acquisition de langues étrangères (anglais, français, mandarin) et l’apprentissage du contrôle chironomique des contours d'intonation de la langue maternelle, pour la suppléance vocale (larynx artificiel). Méthodologie : Les théories phonologiques et phonétiques de l’organisation temporelle des langues étudiées seront considérées dans le contexte du paradigme de la synthèse performative. L’étude des relations entre points de contrôle rythmique, centres perceptifs, gestes de tapotage et unités phonologiques implique la modélisation et l’expérimentation, avec des sujets réalisant des tâches de perceptionaction. La méthodologie relève ici de la psychologie et de la phonétique expérimentales : définition de corpus, mise en oeuvre de protocoles de test, tests, analyses statistiques. Un synthétiseur qui utilise les nouveaux paradigmes de contrôle rythmique sera développé. La méthodologie relève ici du traitement du signal audio et de la parole ainsi que de l’informatique, depuis la conception jusqu’à la programmation. Ainsi un ensemble de méthodes pour le contrôle gestuel du rythme prosodique et du temps sera développé et testé dans les tâches applicatives visées. Ces méthodes comprennent à la fois les gestes et les interfaces de contrôle et relèvent de l’informatique dans le domaine des interfaces humainmachine. Prérequis : Ce sujet est à l’interface de la synthèse vocale et des interfaces humain-machine, de la prosodie, de la perception et de la performance musicale. Cela demande des connaissances générales en traitement du signal audionumérique et en informatique musicale ou en interface humain-machine. Une partie du travail portera sur le développement logiciel. Des connaissances sur la voix et la parole, en phonétique et phonologie, ainsi qu’en psychologie expérimentale ou sciences cognitives seront nécessaires. Les candidatures avec une formation initiale en informatique et traitement du signal aussi bien que celles avec une formation initiale en linguistique, phonétique ou sciences cognitives seront considérées. La formation initiale sera éventuellement complétée dans les domaines qui seraient moins connus. Encadrement : Christophe d’Alessandro, DR CNRS, Responsable de l’équipe LAM Institut Jean Le Rond d’Alembert, Sorbonne Université christophe.dalessandro@sorbonne-universite.fr Ce projet doctoral est dans le cadre du contrat ANR Gepeto, en collaboration avec le LPP (Sorbonne nouvelle), et le GIPSA-Lab, Université de Grenoble. Début du contrat dès que possible (à partir d’octobre 2020) Références : • Delalez, S. et d’Alessandro, C. (2017). “Vokinesis: syllabic control points for performative singing synthesis”, NIME’17 , pp. 198-203. • X. Xiao, G. Locqueville, C. d'Alessandro, B. Doval, « T-Voks: Controlling Singing and Speaking Synthesis with the Theremin », Proceedings of the International Conference on New Interfaces for Musical Expression, NIME’19, June 3-6, 2019, Porto Alegre, Brazil, 110-115. • Samuel Delalez, Christophe d’Alessandro « Adjusting the Frame: Biphasic Performative Control of Speech Rhythm », Proc. INTERSPEECH 2017, 18th Annual Conference of the International Speech Communication Association, Stockholm, Sweden, August 18-25, 2017, DOI: 10.21437/Interspeech.2017, 864-868. • Christophe d’Alessandro, Albert Rilliard, and Sylvain Le Beux « Chironomic stylization of intonation » J. Acoust. Soc. Am., 129(3), march 2011, 1594-1604 • Christophe d’Alessandro, Lionel Feugère, Sylvain Le Beux, Olivier Perrotin, and Albert Rilliard (2014) , « Drawing melodies : evaluation of chironomic singing synthesis » , J. Acoust. Soc. Am. 135 (6), 3601-3612. • I . Chow, M. Belyk, V. Tran, and S. Brown. Syllable synchronisation and the P-center in cantonese. 49 :55–66, 2015. • C. d’Alessandro, L. Feugère, S. Le Beux, and O. Perrotin. Drawing melodies : Evaluation of chironomic singing synthesis. J. Acoust. Soc. Am., 135(6) :3601–3612, March 2014. • C. d’Alessandro, A. Rilliard, and S. Le Beux. Chironomic stylisation of intonation. J. Acoust. Soc. Am., 129(3) :1594–1604, March 2011. • C. Fowler. “Perceptual centers” in speech production and perception. Perception & Psychophysics, 25 :375–388, 1979. • P. Howell. Predicton of p-center location from the distribution of energy in the ampitude envelope. Perception and Psychophysics, 43 :90–93, 1988. • P. F. MacNeilage. The frame/content theory of evolution of speech production. Behavioral and Brain Sciences, 21 :499–546, 1998. • S.M. Marcus. Acoustic determinants of perceptual center (P-center). Perception and Psychophysics, 30 :247–256, 1981. • J. Morton, S. Marcus, and C. Frankish. Perceptual centers (P-Centers). Psychological Review, 83(5) :405–408, 1976. • B. Pompino-Marshall. On the psycho-acoustic nature of the p-center phenomenon. Journal of phonetics, 17 :175–192, 1989. • K. Rapp-Holmgren. A study of syllable timing. STL-QPSR, 12(1) :014–019, 1971. • B. H. Repp. Sensorimtor synchronization : A review of tapping litterature. Psychon. Bull. Rev., 12(6) :969–992, 2005. • B. H. Repp and Y. H. Su. Sensorimotor synchronisation : A review of recent research. Pyschon. Bull. Rev., 20 :403–452, 2013. • P. Wagner. The Rhythm of Language and Speech : Constraining Factors, Models, Metrics and Applications. Habilitation à diriger des recherches, Rheinischen Friedrich-Wilhelms- Universität Bonn, 2008.
| |||||
6-6 | (2020-09-24) Offre de thèse: Modèles profonds pour la reconnaissance et l'analyse de la parole spontanée,Grenoble, France *Sujet*
| |||||
6-7 | (2020-10-05) Research Assistant/Associate in Spoken Language Processing, Cambridge University, UK Research Assistant/Associate in Spoken Language Processing x 2 (Fixed Term)
Speech Research Group, Cambridge University Engineering Department, UK
| |||||
6-8 | (2020-10-05) 2 POsitions at Radbout University, Nijmegen, The Netherlands
We have two vacancies for speech technology employees:
| |||||
6-9 | (2020-10-08) Job offer: 1-year postdoc position at LSP (ENS Paris), France Job offer: 1-year postdoc position at LSP (with a possibility of 1-year extension) The Laboratoire des Systèmes Perceptifs (LSP, ENS Paris / CNRS, https://lsp.dec.ens.fr/en) is offering a postdoc position for the ANR project fastACI ('Exploring phoneme representations and their adaptability using fast Auditory Classification Images') supervised by Léo Varnet (leo.varnet@ens.psl.eu). The fastACI project aims to develop a fast and robust experimental method to visualize and characterize the auditory mechanisms involved in phoneme recognition. The fast-ACI method relies on a stimulus-response model, combined with a reverse correlation (revcorr) experimental paradigm. This approach, producing an 'instant picture' of a participant’s listening strategy in a given context, has already yielded conclusive results, but remains very time-consuming. Therefore, the main objectives of this postdoc contract will be (1) to improve the efficiency of the process using advanced supervised learning techniques (e.g. sparse priors on a smooth basis) and an online adaptive protocol (e.g. Bayesian optimisation); then (2) to use this technique to map the phonemic representations used by normal-hearing listeners. For this purpose, a large number of experiments on phoneme-in-noise categorization tasks (e.g. /aba/ vs. /ada/) will be carried out, in order to insure the broadest possible mapping of the French phonological inventory. As a second step, the new tool will be used to explore the adaptability of speech comprehension in the case of sensorineural hearing loss and noisy backgrounds. The post-doc will be involved in all activities in line with the project, including data collection, coding and statistical analyses. The post-doc will also coordinate trainees’ and students’ work involved in the project, and contribute significantly to publication of the findings. Required profile: Background in psychoacoustics and/or machine learning (the candidate should preferably hold a PhD in one of these fields) High skills in statistical data processing (in particular supervised learning algorithms) and practical knowledge in psychophysics Basic understanding of psychoacoustics and psycholinguistics
Good knowledge of Matlab programming (other languages such as R can also be useful) Strong communication skills in English (working language in the lab) and French (interactions with participants, etc.). Duration: 12 months or 24 months Start: Early 2021 Net salary: ~ 2100€/ month Application Procedure: Applications must include a detailed CV and a motivation letter, a link to (or copy of) the PhD thesis, PhD Viva report, plus the email contact of 2 referees. Applications are to be sent to: Léo Varnet (leo.varnet@ens.psl.eu) before 08/11 (interviews should take place on the 18/11 by videoconference)
| |||||
6-10 | (2020-10-20) PhD grant at INRIA Nancy, France Privacy preserving and personalized transformations for speech recognition
This research position 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. One objective of the project is to transform speech data in order to hide some speaker characteristics (such as voice identity, gender information, emotion, ...) in order to safely share the transformed data while keeping speaker privacy. The shared data is to be used to train and optimize models for speech recognition. The selected candidate will collaborate with other members of the project, and will participate to the project meetings.
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 private [1]. Missions: (objectives, approach, etc.) Within this context, the objective is twofold. First, it aims at improving privacy preserving transforms 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 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. On this topic, recent experiments have relied on Generative Adversarial Networks (GANs) for proposing a privacy preserving transform [Srivastava et al., 2019], and on voice conversion approaches [Srivastava et al., 2020]. In addition, as devices are getting more and more personal, this creates opportunities to make speech recognition more personalized. Some recent studies have investigated approaches that takes benefit of speaker information [Turan et al., 2020]. The candidate will investigate further approaches along these lines. Other topics such as investigating the impact and benefit of adding some random noise in the transforms will be part of the studies, as well as dealing with (hiding) some paralinguistic characteristics. Research directions and priorities will take into account new state-of-the-art results and on-going activities in the project.
Skills and profile: PhD or 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: [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. [Srivastava et al., 2019] B. Srivastava, A. Bellet, M. Tommasi, and E. Vincent: Privacy preserving adversarial representation learning in ASR: reality or illusion? INTERSPEECH 2019 - 20th Annual Conference of the International Speech Communication Association , Sep 2019, Graz, Austria. [Srivastava et al., 2020] B. Srivastava, N. Tomashenko, X. Wang, E. Vincent, J. Yamagishi, M. Maouche, A. Bellet, and M. Tommasi: Design choices for x-vector based speaker anonymization. INTERSPEECH 2020, 21th Annual Conference of the International Speech Communication Association, Oct 2020, Shanghai, China. [Turan et al., 2020] T. Turan, E. Vincent, and D. Jouvet: Achieving multi-accent ASR via unsupervised acoustic model adaptation. INTERSPEECH 2020, 21th Annual Conference of the International Speech Communication Association, Oct 2020, Shanghai, China. [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@inria.fr; https://members.loria.fr/DJouvet/) Duration: 2 years Starting date: autumn 2020 Location: Inria Nancy – Grand Est, 54600 Villers-lès-Nancy
footnote [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 project.
| |||||
6-11 | (2020-10-21) Fully funded PhD position at >IDIAP, Martigny, Switzerland There is a fully funded PhD position open at Idiap Research Institute on 'Speech
| |||||
6-12 | (2020-10-23) PhD grant at Université Grenoble-Alpes, France. Dans le cadre de la Chaire ' Bayesian Cognition and Machine Learning for Speech
| |||||
6-13 | (2020-10-26) Post doctoral position at the Beckman Institute for Advance Science and Technology, University of Illinois, Urbana-Chanpaign, USA Postdoctoralposition in Mobile Sensing and Child Mental Health Beckman Institute for Advanced Science & Technology University ofIllinois at Urbana-Champaign
Our interdisciplinary research team at the Beckman Institute for Advance Science and Technology is developing and applying innovative tools and methods from mobile sensing, signal processing and machine learning to gain insight into the dynamic processes underlying the emergence of disturbances in child mental health. We have engineered a wearable sensing platform that captures speech, motion, and physiological signals of infants and young children in their natural environments, and we are applying data-driven machine-learning approaches and dynamic statistical modeling techniques to large-scale, naturalistic, and longitudinal data sets to characterize dynamic child-parent transactions and children’s developing stress regulatory capacities and to ultimately capture reliable biomarkers of child mental health disturbance.
We seek outstanding candidates for a postdoctoral scholar position that combines multimodal sensing and signal processing, dynamic systems modeling, and child mental health. The ideal candidate would have expertise in one or more of the following domains related to wearable sensors:
In addition to joining a highly interdisciplinary team and making contributions to high impact research on mobile sensing and child mental health, this position provides strong opportunities for professional development and mentorship by faculty team leaders, including Drs. Mark Hasegawa-Johnson, Romit Roy Choudhury, and Nancy McElwain. In collaboration with the larger team, the postdoctoral scholar will play a central role in preparing conference papers and manuscripts for publication, contributing to the preparation of future grant proposals, and assisting with further development of our mobile sensing platform for use with infants and young children.
Applicants should have a doctoral degree in computer engineering, computer science, or a field related to data analytics of wearable sensors, as well as excellent skills in programming, communication, and writing. Appointment is for at least two years, contingent on first-year performance. The position start date is negotiable.
Please send a coverletter and CV to Drs. Mark Hasegawa-Johnson (jhasegaw@illinois.edu) and Nancy McElwain (mcelwn@illinois.edu). Applications will be considered until the position is filled, with priority given to applications submitted by November 15th.
The University of Illinois is an Equal Opportunity, Affirmative Action employer. Minorities, women, veterans and individuals with disabilities are encouraged to apply. For more information, visit http://go.illinois.edu/EEO.
| |||||
6-14 | (2020-10-28) TWO positions in Trinity College Dublin, Ireland we have openings for TWO positions in Trinity College Dublin, Ireland, available from Dec 1st, 2020 for 14 months. We are seeking:
A Research Assistant (qualified to Masters level) A Research Fellow (holds a PhD)
The Project: RoomReader is a project led by Prof. Naomi Harte in TCD and Prof. Ben Cowan in UCD, Ireland. The research is exploring and modelling online interactions, and is funded by the Science Foundation Ireland Covid 19 Rapid Response Call. The candidate will be working with a team to drive research into multimodal cues of engagement in online teaching scenarios. The work involves a collaboration with Microsoft Research Cambridge, and Microsoft Ireland. The Research Assistant will have a psychology/linguistics/engineering background (we are flexible) and will be tasked with researching and designing a new online task to elicit speech based interactions relevant to online teaching scenarios (think multi-party MapTask or Diapix, but different). They will also be responsible for the capture of that dataset and subsequent editing/labelling for deployment in the project and eventual sharing with the wider research community. Annual gross salary up to ?34,930 per annum depending on experience. The Research Fellow needs a background, including a PhD, in deep learning and the modelling of multimodal cues in speech. Their previous experience might be in conversational analysis, multimodal speech recognition or other areas. They should have a proved track record with publications commensurate with career stage. Annual gross salary up to ?50030 depending on experience.
The project starts on Dec 1st, and the positions can start from that date and continue for 14 months. Please email nharte@tcd.ie for a more detailed description of either role, or to discuss. I am open to a person remote-working for the remainder of 2020, but the ideal candidate will be in a position to move to Ireland for Jan 2021 and work with the team in TCD.
Sigmedia Research Group @ Trinity College Dublin, Ireland The Signal Processing and Media Applications (aka Sigmedia) Group was founded in 1998 in Trinity College Dublin, Ireland. Originally with a focus on video and image processing, the group today spans research in areas across all aspects of media ? video, images, speech and audio. Prof. Naomi Harte leads the Sigmedia research endeavours in human speech communication. The group has active research in audio-visual speech recognition, evaluation of speech synthesis, multimodal cues in human conversation, and birdsong analysis. The group is interested in all aspect of human interaction, centred on speech. Much of our work is underpinned by signal processing and machine learning, but we also have researchers with a background in linguistic and psychology aspects of speech processing to keep us all grounded.
| |||||
6-15 | (2020-10-30) 3 Tenure Track Professors (W2) at Saarland University, Germany 3 Tenure Track Professors (W2) at Saarland University
| |||||
6-16 | (2020-10-30) 2 Research Assistant/Associate Posts at Cambridge University, Great Britain 2 Research Assistant/Associate Posts in Spoken Language Processing at Cambridge University (Fixed Term)
Applications are invited for two Research Assistants/Research Associates in the Department of Engineering, Cambridge University, to work on an EPSRC funded project Multimodal Video Search by Examples (MVSE). The project is a collaboration between three Universities, Ulster, Surrey and Cambridge, and the BBC as an industrial partner. The overall aim of the project is to enable effective and efficient multimodal video search of large archives, such as BBC TV programmes.
The research associated with these positions will focus on deriving representation for all the information that is contained within the video speech signal, and integrating with other modalities. The forms of representation will include both: voice analytics e.g. speaker and emotion; and topic and audio content analytics e.g. word-sequence and topic classification and tracking. The position will involve close collaboration with Surrey and Ulster Universities to integrate with other information sources, video and the audio scene, to yield a flexible and efficient video search index.
Fixed-term: The funds for this post are available until 31 January 2024 in the first instance
Closing date: 1st December 2020
Full information can be found at: http://www.jobs.cam.ac.uk/job/27458/
| |||||
6-17 | (2020-11-02) Fully-funded PhD studentships at the University of Sheffield, Great Britain Fully-funded PhD studentships in Speech and NLP at the University of Sheffield *******************************************************************************************************
UKRI Centre for Doctoral Training (CDT) in Speech and Language Technologies (SLT) and their Applications
Department of Computer Science Faculty of Engineering University of Sheffield, UK
Fully-funded 4-year PhD studentships for research in speech technologies and NLP
** Applications now open for September 2021 intake **
Deadline for applications: 31 January 2021.
Speech and Language Technologies (SLTs) are a range of Artificial Intelligence (AI) approaches which allow computer programs or electronic devices to analyse, produce, modify or respond to human texts and speech. SLTs are underpinned by a number of fundamental research fields including natural language processing (NLP / NLProc), speech processing, computational linguistics, mathematics, machine learning, physics, psychology, computer science, and acoustics. SLTs are now established as core scientific/engineering disciplines within AI and have grown into a world-wide multi-billion dollar industry.
Located in the Department of Computer Science at the University of Sheffield ? a world leading research institution in the SLT field ? the UKRI Centre for Doctoral Training (CDT) in Speech and Language Technologies and their Applications is a vibrant research centre that also provides training in engineering skills, leadership, ethics, innovation, entrepreneurship, and responsibility to society.
Apply now: https://slt-cdt.ac.uk/apply/
The benefits:
About you: We are looking for students from a wide range of backgrounds interested in speech and NLP.
Applying: Applications are now sought for the September 2021 intake. The deadline is 31 January 2021.
Applications will be reviewed within 6 weeks of the deadline and short-listed applicants will be invited to interview. Interviews will be held in Sheffield or via videoconference.
See our website for full details and guidance on how to apply: https://slt-cdt.ac.uk
For an informal discussion about your application please contact us by email at: sltcdt-enquiries@sheffield.ac.uk
| |||||
6-18 | (2020-11-14) Master2 Internship, LORIA-INRIA, Nancy, France Master2 Internship: Semantic information from the past in a speech recognition system: does the past help the present?
Supervisor: Irina Illina, MdC, HDR Team and Laboratory: Multispeech, LORIA-INRIA Contact: illina@loria.fr
Co-Supervisor: Dominique Fohr, CR CNRS Team and Laboratory: Multispeech, LORIA-INRIA Contact : dominique.fohr@loria.fr
Motivation and context
Semantic and thematic spaces are vector spaces used for the representation of words, sentences or textual documents. The corresponding models and methods have a long history in the field of computational linguistics and natural language processing. Almost all models rely on the hypothesis of statistical semantics that states that: statistical schemes of appearance of words (context of a word) can be used to describe the underlying semantics. The most used method to learn these representations is to predict a word using the context in which this word appears [Mikolov et al., 2013, Pennington et al., 2014], and this can be realized with neural networks. These representations have proved their effectiveness for a range of natural language processing tasks. In particular, Mikolov's Skip-gram and CBOW models [Mikolov et al., 2013] and BERT model [Devlin et al., 2019] have become very popular because of their ability to process large amounts of unstructured text data with reduced computing costs. The efficiency and the semantic properties of these representations motivate us to explore these semantic representations for our speech recognition system. Robust automatic speech recognition (ASR) is always a very ambitious goal. Despite constant efforts and some dramatic advances, the ability of a machine to recognize the speech is still far from equaling that of the human being. Current ASR systems see their performance significantly decrease when the conditions under which they were trained and those in which they are used differ. The causes of variability may be related to the acoustic environment, sound capture equipment, microphone change, etc.
Objectives Our speech recognition (ASR) system [Povey et al, 2011] is supplemented by a semantic analysis for detecting the words of the processed sentence that could have been misrecognized and for finding words having a similar pronunciation and matching better the context [Level et al., 2020]. For example, the sentence « Silvio Berlusconi, prince de Milan » can be recognized by the speech recognition system as: « Silvio Berlusconi, prince de mille ans ». A good semantic context representation of the sentence could help to find and correct this error. This semantic analysis re-evaluates (rescores) the N-best transcription hypotheses and can be seen as a form of dynamic adaptation in the case of noisy speech data. A semantic analysis is performed in combining predictive representations using continuous vectors. All our models are The semantic module improves significantly the performance of speech recognition system. But we would like to go beyond the semantic information of the current sentence. Indeed, sometimes the previous sentences could help to understand and to recognize the current sentence. The Master internship will be devoted to the innovative study of the taking into account the past recognized sentences to improve the recognition of the current sentence. Research will be conducted on the combination of semantic information from one or several past sentences with semantic information from current sentence to improve the speech recognition. As deep neural networks (DNNs) can model complex functions and get outstanding performance, they will be used in all our modeling. The performance of the different modules will be evaluated on artificially noisy speech data.
Required skills: background in statistics, natural language processing and computer program skills (Perl, Python). Candidates should email a detailed CV with diploma
Bibliography [Devlin et al., 2019] Devlin, J., Chang, M.-W., Lee, K. and Toutanova K. (2019). BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding, Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers). [Level et al., 2020] Level S., Illina I., Fohr D. (2020). Introduction of semantic model to help speech recognition. International Conference on TEXT, SPEECH and DIALOGUE. [Mikolov et al., 2013] Mikolov, T., Sutskever, I., Chen, T. Corrado, G.S.,and Dean, J. (2013). Distributed representations of words and phrases and their compositionality, in Advances in Neural Information Processing Systems, pp. 3111?3119. [Pennington et al., 2014] Pennington, J., Socher, R., and Manning, C. (2014). Glove: Global vectors for word representation. In Proceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 1532-1543. [Povey et al., 2011] Povey, D., Ghoshal, A., Boulianne, G., Burget, L., Glembek, O., Goel, N., Hannemann, M., Motl?cek, P., Qian, Y., Schwarz, Y., Silovsky, J., Stemmer, G., Vesely, K. (2011). The Kaldi Speech Recognition Toolkit, Proc. ASRU.
| |||||
6-19 | (2020-11-15) PhD at LORIA-INRIA, Nancy, France Title: Multi-task learning for hate speech classification
Research Lab: MULTISPEECH team, LORIA-INRIA, Nancy, France
Supervisors: Irina Illina, Associate Professor, HDR (illina@loria.fr)Ashwin Geet D?Sa, PhD Thesis student (ashwin-geet.dsa@loria.fr) Dominique Fohr, Research scientist CNRS (dominique.fohr@loria.fr) Motivation and context: During the last years, online communication through social media has skyrocketed. Although most people use social media for constructive purposes, few misuse these platforms to spread hate speech. Hate speech is anti-social communicative behavior and targets a minor section of the society based on religion, gender, race, etc. (Delgado and Stefancic, 2014). It often leads to threats, fear, and violence to an individual or a group. Online hate speech is a punishable offense by the law, and the owners of the platform are held accountable for the hate speech posted by its users. Manual moderation of hate speech by humans is often expensive and time-consuming. Therefore, automatic classification techniques have been employed for the detection of hate speech.Recently, deep learning techniques have outperformed the classical machine learning techniques and have become state-of-the-art methodology for hate speech classification (Badjatiya et al., 2017). These methodologies need a large quantity of annotated data for training. The task of annotation is very expensive. To train a powerful deep neural network based classification system, several corpora can be used.Multi-task learning is a deep learning-based methodology (MT-DNN) which has proven to outperform the single-task based deep learning models, especially in the low-resource setting (Worsham and Jugal, 2020; Liu et al., 2019). This methodology jointly learns a model on multiple tasks such as classification, question-answering, etc. Thus, the information learned in one task can benefit other tasks and improves the performance of all the tasks.Existing hate speech corpora are collected from various sources such as Wikipedia, Twitter, etc. Labeling of these corpora can vary greatly. Some corpus creators combine various forms of hate, such as ?abuse?, ?threat?, etc. and collectively annotate the samples as ?hate speech? or ?toxic speech?. Whereas, other authors create more challenging corpora using fine-grained annotations such as ?hate speech?, ?abusive speech?, ?racism?, etc. (Davidson et al., 2017). Furthermore, the definition of hate speech remains unclear, and corpus creators can use different definitions. Thus, a model trained on one corpus cannot be easily used to classify the comments from another corpus. To take advantage of the different available hate speech corpora and to improve the performance of hate classification, we would like to explore the multi-corpus learning by using the methodology of multi-task learning.Objectives: The objective of this work is to improve the existing deep learning hate speech classifier by developing the multi-task learning system using several hate speech corpora during the training. In the MT-DNN model of (Liu et al., 2019), the multi-task learning model consists of a set of task-specific layers on top of shared layers. The shared layers are the bottom layers of the model and are jointly trained on several corpora. The task-specific layers are built on top of the shared layers and each of these layers is trained on a single task. We want to explore this setup for hate speech detection learning. In this case, shared layers will be jointly trained on several corpora of hate speech. Each task-specific layer will be used to learn a specific hate speech task from a specific corpus. For example, one task-specific layer can use very toxic, toxic, neither, healthy and very healthy classification task and the Wikipedia detox corpus (Wulczyn et al.,). Another task-specific layer can use hateful, abusive, normal classification task and the Founta corpus (Founta et al.). Since shared-layers are jointly trained using multiple corpora, it will improve the performance of the task-specific layers, especially for a task with small amount of data.The work plan for the internship is as follows: at the beginning, the intern will conduct a literature survey on the hate speech classification using deep neural networks. Using the hate speech classification baseline system (CNN-based or Bi-LSTM-based), existing in our team, the student will evaluate the performance of this system on several available hate speech corpora. After this, the student will develop a new methodology based on the MT-DNN model for efficient learning. We can use pre-trained BERT model (Devlin et al., 2019) to initialize the shared layers of MT-DNN. The performance of the proposed MT-DNN model will be evaluated and compared to single corpora learning and multi-corpora learning (grouping all corpora together).Required Skills: Background in statistics, natural language processing and computer program skills (Python).Candidates should email a detailed CV with diploma. Bibliography: Badjatiya P., Gupta S., Gupta M., and Varma V. ?Deep learning for hate speech detection in tweets.? In Proceedings of the 26th International Conference on World Wide Web Companion, pp. 759-760, 2017. Davidson T., Warmsley D., Macy M., and Weber I. ?Automated hate speech detection and the problem of offensive language.? arXiv preprint arXiv:1703.04009, 2017. Delgado R., and Stefancic J. ?Hate speech in cyberspace.? Wake Forest L. Rev. 49: 319, 2014. Devlin J., Chang M., Lee K., and Toutanova K. ?BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding.? In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers), pp. 4171-4186. 2019. Founta AM, Djouvas C, Chatzakou D, Leontiadis I, Blackburn J, Stringhini G, Vakali A, Sirivianos M, Kourtellis N. Large scale crowdsourcing and characterization of Twitter abusive behavior. ICWSM. 2018. Liu X., He P., Chen W., and Gao J. ?Multi-Task Deep Neural Networks for Natural Language Understanding.? In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pp. 4487-4496. 2019. Worsham J., and Jugal K. ?Multi-task learning for natural language processing in the 2020s: where are we going?.? Pattern Recognition Letters, 2020. Wulczyn E, Thain N, Dixon L. Ex machina: Personal attacks seen at scale. InProceedings of the 26th International Conference on World Wide Web 2017.
| |||||
6-20 | (2020-11-17) 2 year full-time postdoctoral researcher ar the University of Bordeaux, France The University of Bordeaux invites applications for a 2 year full-time postdoctoral researcher in Automatic Speech Recognition. The position is part of the FVLLMONTI project on efficient speech-to-speech translation on embedded autonomous devices, funded by the European Community.
To apply, please send by email a single PDF file containing a full CV (including publication list), cover letter (describing your personal qualifications, research interests and motivation for applying), evidence for software development experience (active Github/Gitlab profile or similar), two of your key publications, contact information of two referees and academic certificates (PhD, Diploma/Master, Bachelor certificates).
Details on the position are given below:
Job description: Post-doctoral position in Automatic Speech Recognition
Duration: 24 months
Starting date: as early as possible (from March 1st 2021)
Project: European FETPROACT project FVLLMONTI (starts January 2021)
Location: Bordeaux Computer Science Lab. (LaBRI CNRS UMR 5800), Bordeaux, France (Image and Sound team)
Salary: from 2 086,45? to 2 304,88?/month (estimated net salary after taxes, according to experience)
Contact: jean-luc.rouas@labri.fr
Short description:
The applicant will be in charge of developping state-of-the-art Automatic Speech Recognition systems for English and French as well as related Machine Translation systems using Deep Neural Networks. The objective is to provide the exact specifications of the designed systems to the other partners of the project specialized in hardware. Adjustements will have to be made to take into account the hardware constraints (i.e. memory and energy consumption impacting the number of parameters, computation time, ...) while keeping an eye on the performances (WER and BLEU scores). When a satisfactory trade-off is reached, more exploratory work is to be carried out on using emotion/attitude/affect recognition on the speech samples to supply additional information to the translation system.
Context of the project:
The aim of the FVLLMONTI project is to build a lightweight autonomous in-ear device allowing speech-to-speech translation. Today, pocket-talk devices integrate IoT products requiring internet connectivity which, in general, is proven to be energy inefficient. While machine translation (MT) and Natural Language Processing (NLP) performances have greatly improved, an embedded lightweight energy-efficient hardware remains elusive. Existing solutions based on artificial neural networks (NNs) are computation-intensive and energy-hungry requiring server-based implementations, which also raises data protection and privacy concerns. Today, 2D electronic architectures suffer from 'unscalable' interconnect and are thus still far from being able to compete with biological neural systems in terms of real-time information-processing capabilities with comparable energy consumption. Recent advances in materials science, device technology and synaptic architectures have the potential to fill this gap with novel disruptive technologies that go beyond conventional CMOS technology. A promising solution comes from vertical nanowire field-effect transistors (VNWFETs) to unlock the full potential of truly unconventional 3D circuit density and performance.
Role:
The tasks assigned to the Computer Science lab are the design of the Automatic Speech Recognition systems (for French and English) and the Machine Translation (English to French and French to English). Speech synthesis will not be explored in the project but an open-source implementation will be used for demonstration purposes. Both ASR and MT tasks benefit from the use of Transformer architectures over Convolutional (CNNs) or Recurrent (RNNs) neural network architectures. Thus, the role of the applicant will be to design and implement state-of-the-art systems for ASR using Transformer nerworks (e.g. with the ESPNET toolkit) and to assist another post-doctorate for the MT systems. Once the performances reached by these baseline systems are satisfactory, details on the network will be given to our hardware designers partners (e.g. number of layers, value of the parameters, etc.). With the feedback of these partners, adjustements will be made to the network considering the hardware constraints while trying not to degrade the performances too much.
The second part of the project will focus on keeping up with the latest innovations and translating them into hardware specifications. For example, recent research suggest that adding convolutional layers to the transformer architecture (i.e. the 'conformer' network) can help reduce the number of parameters of the model which is critical regarding the memory usage of the hardware system.
Finally, more exploratory work on the detection of social affects (i.e. the vocal expression of the intent of the speaker: 'politeness', 'irony', etc) will be carried out. The additional information gathered using this detection will be added to the translation system for potential usage in the future speech synthesis system.
Required skills:
- PhD in Automatic Speech Recognition (preferred) or Machine Translation using deep neural networks
- Knowledge of most widely used toolboxes/frameworks (tensorflow, pytorch, espnet for example)
- Good programming skills (python)
- Good communication skills (frequent interactions with hardware specialists)
- Interest in hardware design will be a plus
Selected references:
S. Karita et al., 'A Comparative Study on Transformer vs RNN in Speech Applications,' 2019 IEEE Automatic Speech Recognition and Understanding Workshop (ASRU), SG, Singapore, 2019, pp. 449-456, doi: 10.1109/ASRU46091.2019.9003750.
Gulati, Anmol, et al. 'Conformer: Convolution-augmented Transformer for Speech Recognition.' arXiv preprint arXiv:2005.08100 (2020).
Rouas, Jean-Luc, et al. 'Categorisation of spoken social affects in Japanese: human vs. machine.' ICPhS. 2019.
---
![]() ![]() Jean-Luc Rouas
| |||||
6-21 | (2020-11-19) Ingenieur de recherches (CDD) -Laboratoire ALAIA, Toulouse Le laboratoire Commun ALAIA, destiné à l'Apprentissage des langues Assisté par Intelligence Artificielle, recrute un ingénieur de recherche en CDD (12 mois avec prolongation possible). Le travail à réaliser se fera en coordination avec les deux partenaires impliqués dans le LabCom : l'IRIT (Institut de Recherche en Informatique de Toulouse) et la société Archean Technologies et plus particulièrement son pôle R&D Archean Labs (Montauban 82). ALAIA est centré sur l'expression et la compréhension orale d'une langue étrangère (L2). Les missions consisteront à concevoir, développer et intégrer des services innovants basés sur l'analyse des productions des apprenants L2, la détection et la caractérisation d'erreurs allant du niveau phonétique au niveau linguistique. Les compétences attendues portent sur le traitement automatique de la parole et du langage, le machine learning et le développement d'application web.
Les candidatures sont à adresser à Isabelle Ferrané (isabelle.ferrane@irit.fr) et Lionel Fontan (lfontan@archean.tech). N'hésitez pas à nous contacter pour de plus amples informations.
| |||||
6-22 | (2020-11-26) Full Professor at Dublin City University, Ireland Dublin City University is seeking to make a strategic appointment to support existing academic and research leadership in the area of Multimodal Information Systems within the DCU School of Computing, home of the ADAPT Centre for Digital Content Technology.
| |||||
6-23 | (2020-11-28) Master2 Internship, INA, Bry sur Marne, FranceDétection de locuteur·rice actif·ve dans les flux télévisuelsSujet de Stage - M2 informatique ou école d?ingénieur
Mots clés (Fr): Détection de locuteur actif, analyse multimodale, traitement de la parole, détection du visage, apprentissage automatique, apprentissage profond, audiovisuel, humanités numériques, place des femmes dans les médias, indexation automatique, gender equality monitor Keywords (En): Active speaker detection, multimodal analysis, speech processing, face detection, machine learning, deep learning, audiovisual, digital humanities, women in media, automatic indexing, gender equality monitor ContexteL?Institut national de l?audiovisuel (INA) est en charge du Dépôt Légal de la télévision, de la radio et du web médias. À ce titre, l?INA capte en continu 170 chaînes de télévision et stocke plus de 20 millions d?heures de contenu audiovisuel.
Un processus d?indexation, généralement réalisé par des documentalistes, est nécessaire pour décrire les contenus audiovisuels et retrouver des documents au sein de ces grandes collections. Ce travail consiste, entre autres, à référencer les personnes apparaissant dans les programmes, les sujets évoqués, ou encore produire des résumés des documents. Les activités du service de la recherche de l?INA visent à automatiser certains processus d?indexation: soit en automatisant certaines tâches sans valeur ajoutée humaine (segmentation, repérage de noms propres dans l?image, etc.), soit en réalisant des tâches qui ne sont pas faites par les documentalistes (décompte exhaustif du temps de parole).
Le sujet proposé s?inscrit dans le cadre du projet Gender Equality Monitor (GEM), financé par l?Agence nationale de la recherche, qui vise à décrire les différences de représentation existant entre les femmes et les hommes dans les médias. Dans ce cadre, des campagnes d?indexation automatique massives des fonds INA ont permis de créer de nouvelles connaissances en science humaines en se fondant sur le temps de parole, le temps d?exposition visuelle, ou encore le contenu des incrustations texte [Dou19a, Dou19b, Dou20].
L?amélioration des systèmes d?indexation automatique nécessite de mettre au point des bases d?exemples représentatives de la diversité des matériaux traités, utilisées pour l?entraînement et l?évaluation. La constitution de bases d?exemples est un enjeu stratégique pour la conception de systèmes fondés sur des processus d?apprentissage automatique et des stratégies d?automatisation de constitution des bases peuvent être envisagées [Sal14]. ObjectifsLa détection de locuteur·rice actif·ve (DLA) est une tâche d?analyse multimodale qui consiste à analyser une vidéo, déterminer si les mouvements d?un des visages apparaissant à l?écran correspondent au signal de parole contenu dans la piste audio. La conception de système DLA peut-être envisagée à l?aide d?approches non supervisées [Chun16] ou supervisées [Rot20]. La DLA répond à plusieurs problématiques métier rencontrées par l?INA.
L?objectif général du stage consiste à mettre au point un système DLA, l?évaluer par rapport aux implémentations open-source existantes, le déployer sur les fonds INA pour constituer des bases d?exemples permettant d?améliorer les logiciels inaSpeechSegmenter et inaFaceGender. En fonction des résultats obtenus, le stage peut déboucher sur une diffusion open-source du système réalisé et/ou une publication scientifique. Compétences requises
Conditions du stageLe stage se déroulera sur une période de 4 à 6 mois au sein du service de la Recherche de l?Ina et pourra débuter à partir de Janvier 2021. Il aura lieu sur le site Bry2, situé au 18 Avenue des frères Lumière, 94366 Bry-sur-Marne. Le stagiaire sera encadré par David Doukhan, Ingénieur R&D au service de la recherche et coordinateur du projet GEM. ContactLes candidat·e·s intéressé·e·s peuvent contacter David Doukhan (ddoukhan@ina.fr) pour plus d?informations, ou directement adresser par courriel une lettre de candidature incluant un Curriculum Vitae. Bibliographie
[Chun16] Chung, J. S., & Zisserman, A. (2016). Out of time: automated lip sync in the wild. In Asian conference on computer vision (pp. 251-263). Springer, Cham.
[Dou18] Doukhan, D., Carrive, J., Vallet, F., Larcher, A., & Meignier, S. (2018). An open-source speaker gender detection framework for monitoring gender equality. In ICASSP (pp. 5214-5218).
[Dou19a] Doukhan, D. (2019) À la radio et à la télé, les femmes parlent deux fois moins que les hommes. La revue des médias
[Dou19b] Doukhan, D., Rezgui, Z., Poels, G., & Carrive, J. (2019). Estimer automatiquement les différences de représentation existant entre les femmes et les hommes dans les médias.
[Dou20] Doukhan, D., Méadel, C., Coulomb-Gully, M. (2020) En période de coronavirus, la parole d?autorité dans l?info télé reste largement masculine. La revue des médias
[Nag20] Nagrani, A., Chung, J. S., Xie, W., & Zisserman, A. (2020). Voxceleb: Large-scale speaker verification in the wild. Computer Speech & Language, 60, 101027.
[Rot20] Roth, J., Chaudhuri, S., Klejch, O., Marvin, R., Gallagher, A., Kaver, L., & Pantofaru, C. (2020). Ava Active Speaker: An Audio-Visual Dataset for Active Speaker Detection. In ICASSP IEEE.
[Sal14] Salmon, F., & Vallet, F. (2014). An Effortless Way To Create Large-Scale Datasets For Famous Speakers. In LREC (pp. 348-352).
| |||||
6-24 | (2020-12-01) Funded PhD Position at University of Edinburgh, Scotland, UK Funded PhD Position at University of Edinburgh
PhD Position: Automatic Affective Behaviour Monitoring through speech and/or multimodal means while preserving user’s privacy
For details please visit: ………………………………………………………………………………………………………………………………………………………………………. About the Project The Advanced Care Research Centre at the University of Edinburgh is a new £20m interdisciplinary research collaboration aiming to transform later life with person centred integrated care The vision of the ACRC is to play a vital role in addressing the Grand Challenge of ageing by transformational research that will support the functional ability of people in later life so they can contribute to their own welfare for longer. With fresh and diverse thinking across interdisciplinary perspectives our academy students will work to creatively embed deep understanding, data science, artificial intelligence, assistive technologies and robotics into systems of health and social care supporting the independence, dignity and quality-of-life of people living in their own homes and in supported care environments. The ACRC Academy will equip future leaders to drive society’s response to the challenges of later life care provision; a problem which is growing in scale, complexity and urgency. Our alumni will become leaders in across a diverse range of pioneering and influential roles in the public, private and third sectors. Automatic affect recognition technologies can monitor a person’s mood and mental health by processing verbal and non-verbal cues extracted from the person’s speech. However, the speech signal contains biometric and other personal information which can, if improperly handled, threaten the speaker’s privacy. Hence there is a need for automatic inference and monitoring methods that preserve privacy for speech data in terms of collection, training of machine learning models and use of such models in prediction. This project will focus on research, implementation and assessment of solutions for handling of speech data in the user’s own environment while protecting their privacy. We are currently studying the use of speech in healthy ageing and care in combination with IoT/Ambient Intelligence technologies in a large research project. This project will build on our research in this area.
The goals of this PhD project are:
Training outcomes include machine learning methods for inference of mental health status, privacy-preserving machine learning and signal processing, and applications of such methods in elderly care.
| |||||
6-25 | (2020-12-03) 6 months internship at GIPSA-Lab, Grenoble, France Deep learning-based speech coding and synthesis in adverse conditions. Projet : Vokkero 2023 Type : Internship, 6 months, start of 2021 Offre : vogo-bernin-pfe-2 Contact : r.vincent@vogo.fr Keywords : Neural vocoding, deep-learning, speech synthesis, training dataset, normalisation. Résumé : The project consists in evaluating the performances of the LPCNet neural vocoder for speech coding and decoding under adverse conditions (noisy environment, varied speech style, etc.) and in proposing learning techniques to improve the quality of synthesis. 1 L’entreprise VOGO, le Gipsa-lab Vogo is an SME based in Montpellier, south of France : www.vogo-group.com. Vogo is the first Sportech listed on Euronext Growth and develops solutions that enrich the experience of fans and professionals during sporting events. Its brand Vokkero is specialized in the design and production of radio communication systems : www.vokkero.com. It offers solutions for teams working in very noisy environments and is notably a world reference in the professional sports refereeing market. Gipsa-lab is a CNRS research unit joint with Grenoble-INP (Grenoble Institute of Technology), and Université Grenoble Alpes. With 350 people, including about 150 doctoral students, Gipsa-lab is a multidisciplinary research unit developing both basic and applied researches on complex signals and systems. Gipsa-lab is internationally recognised for the research achieved in Automatic Control, Signal and Images processing, Speech and Cognition, and develops projects in the strategic areas of energy, environment, communication, intelligent systems, Life and Health and language engineering. 2 Le projet Vokkero 2023 Every 3 years, Vokkero renews its Hardware (radio, cpu) and Software (rte, audio processing) platforms, in order to design new generations of products. The project extends over several years of study and it is within this framework that the internship is proposed. In the form of a partnership with the Gipsalab, the project consists in the study of speech coding using « neural networks » approaches, in order to obtain performances not yet reached by classical approaches. The student will work at the GIPSA-lab in the CRISSP team of the Speech and Cognition cluster under the supervision of Olivier PERROTIN, research fellow at CNRS, and at the R&D of Vogo Bernin, with Rémy VINCENT, project leader on the Vogo side. 3 Context & Objectives The project consists in evaluating the performances of the LPCNet neural vocoder for speech coding and decoding under adverse conditions (noisy environment, varied speech style, etc.) and in proposing learning techniques to improve the quality of synthesis. 3.1 Context Vocoders (voice coders) are models that allow a speech signal to be first reduced to a small set of parameters (this is speech analysis or coding) and then reconstructed from these parameters (this is speech synthesis or decoding). This coding/decoding process is essential in telecommunication applications, where speech is coded, transmitted and then decoded at the receiver. The challenge is to minimise the quantity of information transmitted, while keeping the quality of the reconstructed speech signal as high as possible. Current techniques use high-quality speech signal models, with a constraint on algorithmic complexity to ensure real-time processes in embedded systems. Examples of Codecs widelay used are Speex (Skype) and its little brother, Opus (Zoom). A few orders of magnitude : OPUS converts a sampled stream at 16kHz into a bitstream at 16kbits (i.e. a compression ratio of 1 :16), the reconstructed signal is also at 16kHz and has 20ms of latency. Since 2016 a new type of vocoder has emerged, called neural vocoder. Based on deep neural network architectures, these are able to generate a speech signal from the classical input parameters of a vocoder, without a priori knowledge of an explicit speech model, but using machine learning. The first system, Google’s WaveNet [1], is capable of reconstructing a signal almost identical to natural speech, but at a very high computation cost (20 seconds to generate a sample, 16,000 samples per second). Since then, models have been simplified and are capable of generating speech in real time (WaveRNN [2], WaveGlow [3]). In particular, the LPCNet neural vocoder [4, 5], also developed by Mozilla, is able to convert a 16kHz sampled stream into a 4kbits bitstream, and reconstruct a 16kHz audio signal. This mix of super-compression combined with bandwidth extension leads to much higher equivalent compression ratios than 1 :16 ! However, the ability of these systems to generate high-quality speech has only been evaluated following training on large and homogeneous databases, i.e. 24 hours of speech read by a single speaker and recorded in a quiet environment [6]. On the other hand, in the application of Vokkero, speech is recorded in adverse conditions (very noisy environment), and presents a significant variability (spoken voice, shouted voice, multiplicity of referees, etc.). Is a neural vocoder trained on a read speech database capable of decoding speech of this type? If not, is it possible to train the model on such data, while they are only available in small quantities ? The aim of this internship is to explore the limits of the LPCNet vocoder in application to the decoding of referee speech. Various learning strategies (curriculum training, transfer learning, learning on augmented data, etc.) will then be explored to try to adapt pre-trained models to our data. 3.2 Tasks The student will evaluate the performance of a pre-trained LPCNet vocoder on referee speech data, and will propose learning strategies to adapt the model to this new data, in a coding/re-synthesis scena rio : 1. Get familiar with the system, performance evaluation on an audio-book database (baseline) ; 2. Evaluation of LPCNet on the Vokkero database and identification of the limits (ambient noise, pretreatments, voice styles, etc.) ; 3. Study of strategies to improve system performance by data augmentation : — Creation of synthetic and specific databases : noisy atmospheres, shouted voices ; — Recording campaigns on Vokkero systems, in anechoic rooms and/or real conditions if the sanitary situation allows it ; — Comparison of the 2 approaches according to various learning strategies to learn a new model from this data. 3.3 Required Skills The student is expected to have a solid background in speech signal processing and an interest in Python development. Experience in programming deep learning models in Python is a plus. The student is expected to show curiosity for research, scientific rigour in methodology and experimentation, and show autonomy for technical and organisational aspects. Depending on the candidate’s motivation, and subject to obtaining funding, it is possible to pursue this topic as a PhD thesis. The student will be able to subscribe to the company’s insurance system, will have luncheon vouchers and will receive a monthly gratuity of 800€. Références [1] A. van den Oord, S. Dieleman, H. Zen, K. Simonyan, O. Vinyals, A. Graves, N. Kalchbrenner, A. W. Senior et K. Kavukcuoglu, “WaveNet : A Generative Model for Raw Audio”, CoRR, t. abs/1609.03499, 2016. arXiv : 1609.03499 (cf. p. 1). [2] N. Kalchbrenner, E. Elsen, K. Simonyan, S. Noury, N. Casagrande, E. Lockhart, F. Stimberg, A. van den Oord, S. Dieleman et K. Kavukcuoglu, “Efficient Neural Audio Synthesis”, CoRR, t. abs/1802.08435, 2018. arXiv : 1802.08435 (cf. p. 1). [3] R. Prenger, R. Valle et B. Catanzaro, “Waveglow : A Flow-based Generative Network for Speech Synthesis”, in Proceedings of the International Conference on Acoustics, Speech and Signal Processing (ICASSP), Brighton, UK : IEEE, mai 2019, p. 3617-3621 (cf. p. 1). [4] J.-M. Valin et J. Skoglund, “LPCNET : Improving Neural Speech Synthesis through Linear Prediction”, in IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP), sér. ICASSP ’19, Brighton, UK : IEEE, mai 2019, p. 5891-5895 (cf. p. 1). [5] ——, “A Real-Time Wideband Neural Vocoder at 1.6kb/s Using LPCNet”, in Proceedings of Interspeech, Graz, Austria : ISCA, sept. 2019, p. 3406-3410 (cf. p. 1). [6] P. Govalkar, J. Fischer, F. Zalkow et C. Dittmar, “A Comparison of Recent Neural Vocoders for Speech Signal Reconstruction”,
| |||||
6-26 | (2020-12-04) 6 mois de post-doctorat, IRIT, Toulouse, France Dans le cadre du projet interdisciplinaire INGPRO, projet Région Occitanie sur l'étude de l'Incidence des Gestes sur la PROnonciation, l'IRIT (équipe SAMoVA https://www.irit.fr/SAMOVA/site/) propose 6 mois de post-doctorat pour travailler sur le traitement des donne?es de parole (évaluation manuelle et automatique) recueillies lors d'une expe?rimentation qui se de?roulera au printemps 2021. Cette expe?rimentation implique le recueil de donne?es orales dans différentes conditions expe?rimentales ainsi que l'analyse des données collectées. Ce travail se fera en collaboration avec la société Archean Technologie (http://www.archean.tech/archean-labs-en.html) et le laboratoire Octogone de l'UT2J (https://octogone.univ-tlse2.fr/), partenaires du projet. Si vous êtes intéressé.e, vous trouverez ci-dessous le détail de l'offre proposée. Offre : https://www.irit.fr/SAMOVA/site/wp-content/uploads/2020/12/Ficheposte_PostDoc_INGPRO_2021.pdf Des compléments sur le projet INGPRO sont accessibles ici : https://www.irit.fr/SAMOVA/site/projects/current/ingpro/
Poste à pourvoir : CDD (post-doc catégorie A)
| |||||
6-27 | (2020-12-05) 5/6 months Internship, LIS-Lab, Université Aix-Marseille, France Deep learning for speech perception
| |||||
6-28 | (2020-12-06) chercheur mi-temps, Université de Mons, Belgique Le laboratoire de phonétique de l?UMONS (https://sharepoint1.umons.ac.be/FR/universite/facultes/fpse/serviceseetr/sc_langage/Pages/default.aspx), rattaché à l?Institut de Recherches en Sciences du Langage (https://langage.be/), est un labo SHS partenaire d?un projet R&D porté par la société https://roomfourzero.com/ et financé par la DG06 Wallonie (https://recherche-technologie.wallonie.be/): le projet « SALV », pour « sens à la voix ».
Pour mener à bien ce projet, nous recrutons un chercheur à mi-temps pour une première période de 6 mois ou à temps plein pour une première période de 3 mois à partir de janvier 2021 (avec possibilité d?extension à 1 ou 2 ans).
La mission est double :
Les modalités pratiques de l?engagement sont à ce stade très flexibles, à discuter avec les possibilités/souhaits du candidat retenu. Le salaire est fixé selon le diplôme (idéalement : doctorat en informatique ; autres formations : à discuter) et l?expérience.
Si vous êtes intéressé.e ou si vous voulez simplement en savoir plus : veronique.delvaux@umons.ac.be
**************** SALV
Room 40 (https://roomfourzero.com/) est une société jeune et dynamique qui fournit un ensemble de produits et services incluant notamment l?analyse en temps réel et la détection d?anomalie dans des flux audios et vidéos, ainsi que l?analyse contextuelle fine, basée sur le concept d?ontologie, des significations explicites et implicites de textes et bribes de textes tels que ceux échangés sur les réseaux sociaux ou via SMS.
Le projet SALV a pour objectif de développer une technologie de transcription et d?analyse de contenu vocal incluant un ensemble d?informations paralinguistiques contextuelles (émotions, stress, attitudes). Il se base en partie sur l?utilisation de technologies d?analyse de texte en temps réel et d?ontologies spécifiques que Room 40 a sous licence et commercialise déja?. Pour ce projet, de nouveaux outils de transcription de parole en texte devront e?tre développés afin d?intégrer ces informations paralinguistiques sous forme de métadonnées au texte de la transcription. La conjonction de ces deux types d?information devrait grandement améliorer la qualité du résultat de l?analyse.
Au terme du projet, on vise une solution d?analyse de parole comportant: (i) une approche pour retranscrire des paroles en texte; (ii) une méthode pour annoter le texte de métadonnées reprenant des éléments paralinguistiques; (iii) un syste?me pour l?analyse pertinente du contenu audio combinant la retranscription et des éléments paralinguistiques; enfin (iv) une couche applicative intégrant les éléments ci-dessus et comprenant des algorithmes d?analyse de contenu, des interfaces graphiques spécifiques a? certains segments de marché, ainsi qu?un nombre d?APIs garantissant l?interopérabilité du syste?me avec les infrastructures existantes des partenaires.
| |||||
6-29 | (2020-12-07) Internships at IRIT, Toulouse, France L?équipe SAMoVA de l?IRIT à Toulouse propose plusieurs stages de fin d?étude (M2, ingénieur) en 2021 :
| |||||
6-30 | (2020-11-20) 6 month-Internship, Ludo-Vic SAS, Paris France 6 mois de stage Master 2 Détection de baisse d’engagement VUE D'ENSEMBLE L’objectif principal de ce stage est la détection de baisse d’engagement durant une interaction avec nos agents conversationnels. La solution peut être trouvée en utilisant des modèles à base de règles ou en utilisant des techniques de machine/deep learning [1, 2, 3, 4, 5]. OBJECTIFS 1. Analyser le comportement (mouvement de tête, émotion, ...) pour trouver les caractéristiques de baisse d’engagement. 2. Modélisation et détection de la baisse d’engagement 3. Evaluation 4. Application en temps réel Conditions du stage Le stage se déroulera sur une période de 6 mois dans le département R&D du Ludo-Vic SAS. Des outils de travail à distance sont disponibles au sein de l’entreprise. Profil recherché Bac +5 dans le domaine de l’informatique et de l'IA. Capacité à réaliser des interactions et des animations 3D. Expérience avec Unity3D et compétence en langage C# sont un vrai plus. Rémunération : conditions standards de rémunération de stage. CONTACTS ET CANDIDATURE Merci d’envoyer votre CV (vos relevés de notes, vos rapports de projets/stages…) à : - Jack Amberg : jack@ludo-vic.com - Atef Ben-Youssef : atef@ludo-vic.com
| |||||
6-31 | (2020-12-16) Master 2 / PFE internship at GIPSA-lab (Grenoble) Stage MASTER / PFE 2020-2021 REAL-TIME SILENT SPEECH SYNTHESIS BASED ON END-TO-END DEEP LEARNING MODELS Context Various pathologies affect the voice sound source, i.e. the vibration of the vocal folds, thus preventing any sound production despite the normal functioning of articulators (movements of the jaw, tongue, lips, etc.): this is known as silent speech. Silent speech interfaces [Denby et al., 2010] consist in converting inaudible cues such as articulators movements into an audible speech signal to rehabilitate the speaker’s voice. At GIPSA-lab, we have a system for measuring articulators using ultrasound imaging and video and for converting this data into acoustic parameters that describe a speech signal, using machine learning [Hueber and Bailly, 2016, Tatulli and Hueber, 2017]. The speech signal is then reconstructed from the predicted acoustic parameters using a vocoder [Imai et al., 1983]. Current silent speech interfaces have two main limitations: 1) The intonation (or speech melody), normally produced by the vibration of the vocal folds, is absent in the considered type of pathologies and is difficult to reconstruct from articulatory information only; 2) The generated speech quality is often limited by the type of vocoder used. While the recent emergence of neural vocoders has allowed a leap in the quality of speech synthesis [van den Oord et al., 2016], they have not yet been integrated into silent speech interface, where the constraint of real-time generation is crucial. Objectives Mapping We propose in this internship to address these two problems, by implementing an end-to-end silent speech synthesis system with deep learning models. In particular, it will consist in interfacing our system for articulation measurement and acoustic parameter generation with the LPCNet neural vocoder [Valin and Skoglund, 2019]. The latter takes asinput acoustic parameters coming from articulation on the one hand, and the intonation on the other hand. This distinction offers the possibility of decorrelating both controls, by proposing a gestural control of the intonation for example [Perrotin, 2015]. Regarding the acoustic parameters, the first step will be to adapt the acoustic output of our system to match theinput of LPCNet. Moreover, LPCNet is trained by default on acoustic parameters extracted from natural speech, forwhich large databases are available. However, the acoustic parameters predicted from silent speech are degraded, and produced in small quantities. We will thus study the robustness of LPCNet to a degraded input, and several re-training strategies (adaptation of LPCNet to new data, end-to-end learning, etc.) will be explored. Once the system is functional, the second part of the internship will consist in implementing the system in real-time, so that the speech s ynthesis is generated synchronously with the user’s articulation. All stages of implementation (learning strategies, real-time system) will be evaluated in terms of intelligibility, sound quality, and intonation reconstruction. Tasks The tasks expected during this internship are: Implement the full silent speech synthesis pipeline by interfacing the lab ultrasound system with LPCNet, and explore training strategies. Evaluate the performance of the system regarding speech quality and reconstruction errors. Implement and evaluate a real-time version of the system. Required skills Signal processing and machine learning. Knowledge of Python and C is required for implementation. Knowledge of Max/MSP environment would be a plus for real-time implementation. Strong motivation for methodology and experimentation. Allowance The internship allowance is fixed by ministerial decree (about 570 euros / month). Grenoble Images Parole Signal Automatique UMR CNRS 5216 – Grenoble Campus 38400 Saint Martin d’Hères - FRANCE Stage MASTER / PFE 2020-2021 Contact Olivier PERROTIN + 33 4 76 57 45 36 olivier.perrotin@grenoble-inp.fr Thomas HUEBER + 33 4 76 57 49 40 thomas.hueber@grenoble-inp.fr References [Denby et al., 2010] Denby, B., Schultz, T., Honda, K., Hueber, T., Gilbert, J. M., and Brumberg, J. S. (2010). Silent speech interfaces. Speech Communication, 52(4):270–287. [Hueber and Bailly, 2016] Hueber, T. and Bailly, G. (2016). Statistical conversion of silent articulation into audible speech using fullcovariance hmm. Computer Speech & Language, 36(Supplement C):274–293. [Hueber et al., 2010] Hueber, T., Benaroya, E.-L., Chollet, G., Denby, B., Dreyfus, G., and Stone, M. (2010). Development of a silent speech interface driven by ultrasound and optical images of the tongue and lips. Speech Communication, 52(4):288–300. [Imai et al., 1983] Imai, S., Sumita, K., and Furuichi, C. (1983). Mel log spectrum approximation (mlsa) filter for speech synthesis. Electronics and Communications in Japan (Part I: Communications), 66(2):10–18. [Perrotin, 2015] Perrotin, O. (2015). Chanter avec les mains: Interfaces chironomiques pour les instruments de musique numériques. PhD thesis, Université Paris-Sud, Orsay, France. [Tatulli and Hueber, 2017] Tatulli, E. and Hueber, T. (2017). Feature extraction using multimodal convolutional neural networks for visual speech recognition. In Proceedings of the International Conference on Acoustics, Speech and Signal Processing (ICASSP), ICASSP ’17, pages 2971–2975, New Orleans, LA, USA. [Valin and Skoglund, 2019] Valin, J.-M. and Skoglund, J. (2019). Lpcnet: Improving neural speech synthesis through linear prediction. In IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP), ICASSP ’19, pages 5891–5895, Brighton, UK. IEEE. [van den Oord et al., 2016] van den Oord, A., Dieleman, S., Zen, H., Simonyan, K., Vinyals, O., Graves, A., Kalchbrenner, N., Senior, A. W., and Kavukcuoglu, K. (2016). Wavenet: A generative model for raw audio. CoRR, abs/1609.03499.
| |||||
6-32 | (2020-12-18) Ingenieur de recherche (CDD), Lab. ALAIA, France Le laboratoire Commun ALAIA, destiné à l'Apprentissage des langues Assisté par Intelligence Artificielle, recrute un ingénieur de recherche en CDD (12 mois avec prolongation possible). Le travail à réaliser se fera en coordination avec les deux partenaires impliqués dans le LabCom : l'IRIT (Institut de Recherche en Informatique de Toulouse) et la société Archean Technologies et plus particulièrement son pôle R&D Archean Labs (Montauban 82). ALAIA est centré sur l'expression et la compréhension orale d'une langue étrangère (L2). Les missions consisteront à concevoir, développer et intégrer des services innovants basés sur l'analyse des productions des apprenants L2, la détection et la caractérisation d'erreurs allant du niveau phonétique au niveau linguistique. Les compétences attendues portent sur le traitement automatique de la parole et du langage, le machine learning et sont indispensables pour être opérationnel dès la prise de fonction. De bonnes connaissances en développement d'applications web seraient un plus.
Les candidatures sont à adresser à Isabelle Ferrané (isabelle.ferrane@irit.fr) et Lionel Fontan (lfontan@archean.tech). N'hésitez pas à nous contacter pour de plus amples informations.
| |||||
6-33 | (2021-01-03) Postdoc researcher, Seikei University, Japan We are seeking a highly motivated and ambitious post-doctoral researcher for the project of ?socially and culturally aware human-agent interaction,? led by Prof. Yukiko Nakano at Seikei University in Tokyo, Japan. The project is part of a larger government funded project for a human-avatar symbiotic society. The mission of our group (http://iui.ci.seikei.ac.jp/en/) is to research and develop technologies for human-agent/robot interaction, behavior generation, and behavior adaptation by focusing on social and cultural awareness.
| |||||
6-34 | (2021-01-04) Open positions at IDIAP, Martigny, Suisse There is a fully funded PhD position open at Idiap Research Institute on 'Neural
| |||||
6-35 | (2021-01-04) research scientist in spoken language processing at Naver Labs Europe, Grenoble, France We are seeking to recruit a research scientist in spoken language processing at Naver Labs Europe (Grenoble, France) - https://europe.naverlabs.com
More details below (you can apply online here as well)
DESCRIPTIONNAVER LABS Europe's mission is to create new ways to interact with digital and physical agents, while paving the way for these innovations into a number of NAVER flagship products and services. This includes research in models and algorithms to give humans faster and better access to data and to allow them to interact with technology in simpler and more natural ways. To fulfill our vision of intelligent devices communicating seamlessly with us, we need to considerably improve existing technology and methods that solve natural language processing problems. We are looking for applications from research scientists to make outstanding contributions to the invention, development and benchmarking of spoken language processing techniques. The research scientist would be part of the Natural Language Processing group of NAVER LABS Europe and her mission would be to develop research on one or more of the following themes: spoken language translation, speech recognition, text-to-speech synthesis, voice-based conversational search (with potential collaborations with the Search&Recommendation group). At NAVER LABS we encourage participation in the academic community. Our researchers collaborate closely with universities and regularly publish in venues such as ACL, EMNLP, Interspeech, KDD, SIGIR, ICLR, ICML and NeurIPS. REQUIRED SKILLS- Ph.D. in spoken language processing, speech processing, NLP or machine learning.
- Knowledge of latest developments in statistical and deep learning as applied to NLP and speech. - Strong publication record in top-tier NLP, speech or machine learning conferences. - Strong development skills, preferably in python and knowledge of relevant frameworks (tensorflow, pytorch, etc). APPLICATION INSTRUCTIONSYou can apply for this position online. Don't forget to upload your CV and cover letter before you submit. Incomplete applications will not be accepted.
ABOUT NAVER LABSNAVER LABS Europe has full-time positions, PhD and PostDoc opportunities throughout the year which are advertised here and on international conference sites that we sponsor such as CVPR, ICCV, ICML, NeurIPS, EMNLP, ACL etc. NAVER LABS Europe is an equal opportunity employer. NAVER LABS are in Grenoble in the French Alps. We have a multi and interdisciplinary approach to research with scientists in machine learning, computer vision, artificial intelligence, natural language processing, ethnography and UX working together to create next generation technology and services that deeply understand users and their contexts.
| |||||
6-36 | (2021-01-07) Speech-NLP Master 2 Internship Year 2020-2021 at LISN (ex LIMSI), University Paris-Saclay, France Speech-NLP Master 2 Internship Year 2020-2021 Speech Segmentation and Automatic Detection of Conflicts in Political Interviews LISN – Université Paris-Saclay Internship for Last Year Engineer or Master 2 Students Keywords: Machine Learning, Diarization, Digital Humanities, Political Speech, Prosody, Expressive Speech Context This internship is part of the Ontology and Tools for the Annotation of Political Speech (OOPAIP 2018), a transdisciplinary project funded under the DIM-STCN (Text Sciences and New Knowledge) by the Regional Council of Ile de France. The project is carried out by the European Center for Sociology and Political Science (CESSP) of the University of Paris 1 Panthéon-Sorbonne, the National Audiovisual Institute (INA), and the LISN. Its objective is to design new approaches to develop detailed, qualitative, and quantitative analyzes of political speech in the French media. Part of the project concerns the study of the dynamics of conflicting interactions in interviews and political debates, which requires a detailed description and a large corpus to allow for the models’ generalization. Some of the main challenges concern the performance of speaker and speech style segmentation, e.g., improving the segmentation accuracy, detecting superimposed speech, measuring vocal effort and other expressive elements. Objectives The main objective of the internship is to improve the automatic segmentation of political interviews. In this context, we will be particularly interested in the detection of hubbub (strong and prolonged overlapped speech). More precisely, we would like to extract features from the speech signal (Eyben et al. 2015) correlated with the level of conflictual content in the exchanges, based, for example, on the arousal level in the speaker’s voice—intermediate level between the speech signal analysis and the expressivity description (Rilliard, d’Alessandro, and Evrard 2018)—or vocal effort (Liénard 2019). The internship will initially be based on two corpora of 30 political interviews manually annotated in speech turns and speech acts—within the framework of the OOPAIP project. It will begin with a state of the art review of speech diarization and overlapped speech detection (Chowdhury et al. 2019). The aim will then be to propose solutions based on recent frameworks (Bredin et al. 2020) to improve the precise localization of speaking segments, in particular when the frequency of speaker changes is high. In the second part of the internship, we will look at a more detailed measurement and prediction of the conflicting level of exchanges. We will search for the most relevant features to describe the conflicting level and by adapting or developing a neural network architecture for its modeling. The programming language used for this internship will be Python. The candidate will have access to the LISN computing resources (servers and clusters with recent generation GPUs).
Publications Depending on the degree of maturity of the work carried out, we expect the applicant to: • Distribute the tools produced under an open-source license • Write a scientific publication Conditions The internship will take place over a period of 4 to 6 months at the LISN (formerly LIMSI) in the TLP group (spoken language processing). The laboratory is located near the plateau de Saclay, university campus building 507, rue du Belvédère, 91400 Orsay. The candidate will be supervised by Marc Evrard (evrard@limsi.fr). Allowance under the official standards (service-public.fr). Applicant profile • Student in the last year of a 5-years diploma in the field of computer science (AI is a plus) • Proficiency in Python language and experience in using ML libraries (Scikit-Learn, Tensor- Flow, PyTorch) • Strong interest in digital humanities and political science in particular • Experience in automatic speech processing is preferred • Ability to carry out a bibliographic study from scientific articles written in English To apply: Send an email to evrard@limsi.fr including a résumé and a cover letter. Bibliography Bredin, Hervé, Ruiqing Yin, Juan Manuel Coria, Gregory Gelly, Pavel Korshunov, Marvin Lavechin, Diego Fustes, Hadrien Titeux, Wassim Bouaziz, and Marie-Philippe Gill. 2020. “Pyannote. Audio: Neural Building Blocks for Speaker Diarization.” In ICASSP. IEEE. Chowdhury, Shammur Absar, Evgeny A Stepanov, Morena Danieli, and Giuseppe Riccardi. 2019. “Automatic Classification of Speech Overlaps: Feature Representation and Algorithms.” Computer Speech & Language 55: 145–67. Eyben, Florian, Klaus R Scherer, Björn W Schuller, Johan Sundberg, Elisabeth André, Carlos Busso, Laurence Y Devillers, et al. 2015. “The Geneva Minimalistic Acoustic Parameter Set (GeMAPS) for Voice Research and Affective Computing.” IEEE Transactions on Affective Computing 7 (2): 190–202. Liénard, Jean-Sylvain. 2019. “Quantifying Vocal Effort from the Shape of the One-Third Octave Long-Term-Average Spectrum of Speech.” The Journal of the Acoustical Society of America 146 (4): EL369–75. OOPAIP. 2018. “(Ontologie Et Outil Pour l’annotation Des Interventions Politiques).” DIM STCN (Sciences du Texte et connaissances nouvelles) Conseil régional d’Ile de France. http://www.dim-humanites-numeriques.fr/projets/oopaip-ontologie-et-outilspour- lannotation-des-interventions-politiques/. Rilliard, Albert, Christophe d’Alessandro, and Marc Evrard. 2018. “Paradigmatic Variation of Vowels in Expressive Speech: Acoustic Description and Dimensional Analysis.” The Journal of the Acoustical Society of America 143 (1): 109–22.
|