ISCA - International Speech
Communication Association

ISCApad Archive  »  2020  »  ISCApad #267  »  Resources

ISCApad #267

Thursday, September 10, 2020 by Chris Wellekens

5 Resources
5-1 Books
5-1-1Emmanuel Vincent (Editor), Tuomas Virtanen (Editor), Sharon Gannot (Editor), 'Audio Source Separation and Speech Enhancement', Wiley

 Emmanuel Vincent (Editor), Tuomas Virtanen (Editor), Sharon Gannot (Editor),

Audio Source Separation and Speech Enhancement:

ISBN: 978-1-119-27989-1

October 2018

504 pages

This 500-page book provides a unifying view of source separation and enhancement,
including but not limited to array processing, matrix factorization, and deep learning
based methods, and speech and music applications, with consistent notation and
terminology across all chapters.


5-1-2Jen-Tzung Chien, 'Source Separation and Machine Learning', Academic Press

Jen-Tzung Chien, 'Source Separation and Machine Learning', Academic Press

Source Separation and Machine Learning presents the fundamentals in adaptive learning
algorithms for Blind Source Separation (BSS) and emphasizes the importance of machine
learning perspectives. It illustrates how BSS problems are tackled through adaptive
learning algorithms and model-based approaches using the latest information on mixture
signals to build a BSS model that is seen as a statistical model for a whole system.
Looking at different models, including independent component analysis (ICA), nonnegative
matrix factorization (NMF), nonnegative tensor factorization (NTF), and deep neural
network (DNN), the book addresses how they have evolved to deal with multichannel and
singlechannel source separation.

Key features:
? Emphasizes the modern model-based Blind Source Separation (BSS) which closely connects
the latest research topics of BSS and Machine Learning
? Includes coverage of Bayesian learning, sparse learning, online learning,
discriminative learning and deep learning
? Presents a number of case studies of model-based BSS, using a variety of learning
algorithms that provide solutions for the construction of BSS systems


5-1-3Ingo Feldhausen, « Methods in prosody: A Romance language perspective », Language Science Press (open access)

Nous sommes heureux de vous annoncer la parution d'un recueil validé par un comité de lecture et consacré aux méthodes de recherche en prosodie. Cet ouvrage est intitulé « Methods in prosody: A Romance language perspective ».

Il est publié par Language Science Press, une maison d’édition open access. Le livre peut-être téléchargé gratuitement en cliquant sur le lien suivant :

La table des matières est la suivante :


Ingo Feldhausen, Jan Fliessbach & Maria del Mar Vanrell                                                                   iii

Pilar Prieto                                                                                                                                              vii

I Large corpora and spontaneous speech

1) Using large corpora and computational tools to describe prosody: An
exciting challenge for the future with some (important) pending problems to solve

Juan María Garrido Almiñana                                                                                                                  3

2) Intonation of pronominal subjects in Porteño Spanish: Analysis of 
spontaneous speech

Andrea Pešková                                                                                                                                     45

II Approaches to prosodic analysis

3) Multimodal analyses of audio-visual information: Some methods and
issues in prosody research

Barbara Gili Fivela                                                                                                                                 83

4) The realizational coefficient: Devising a method for empirically
determining prominent positions in Conchucos Quechua

Timo Buchholz & Uli Reich                                                                                                                 123

5) On the role of prosody in disambiguating wh-exclamatives and
wh-interrogatives in Cosenza Italian

Olga Kellert, Daniele Panizza & Caterina Petrone                                                                               165

III Elicitation methods

6) The Discourse Completion Task in Romance prosody research: Status
quo and outlook

Maria del Mar Vanrell, Ingo Feldhausen & Lluïsa Astruc                                                                    191

7) Describing the intonation of speech acts in Brazilian Portuguese:
Methodological aspects

João Antônio de Moraes & Albert Rilliard                                                                                           229

Indexes                                                                                                                                                  263


N'hésitez pas à diffuser la parution de cet ouvrage auprès de vos collègues qui pourraient s'y intéresser.

Bien cordialement,

Ingo Feldhausen
(Co-coordinateur d'ouvrage)


5-1-4Nigel Ward, 'Prosodic Patterns in English Conversation', Cambridge University Press, 2019

Prosodic Patterns in English Conversation

Nigel G. Ward, Professor of Computer Science, University of Texas at El Paso

Cambridge University Press, 2019.


Spoken language is more than words: it includes the prosodic features and patterns that speakers use, subconsciously, to frame meanings and achieve interactional goals. Thanks to the application of simple processing techniques to spoken dialog corpora, this book goes beyond intonation to describe how pitch, timing, intensity and voicing properties combine to form meaningful temporal configurations: prosodic constructions. Combining new findings with hitherto-scattered observations from diverse research traditions, this book enumerates twenty of the principal prosodic constructions of English.   


5-1-5J.H.Esling, Scott R.Moisik, Allison Benner, Lise Crevier-Buchman, 'Voice Quality: the Laryngeal Articulator Model', Cambridge University Press

Voice Quality

The Laryngeal Articulator Model

Hardback 978-1-108-49842-5

John H. Esling, University of Victoria, British Columbia

Scott R. Moisik, Nanyang Technological University, Singapore

Allison Benner, University of Victoria, British Columbia

Lise Crevier-Buchman, Centre National de la Recherche Scientifique (CNRS), Paris

The first description of voice quality production in forty years, this book

provides a new framework for its study: The Laryngeal Articulator Model.

Informed by instrumental examinations of the laryngeal articulatory

mechanism, it revises our understanding of articulatory postures to explain

the actions, vibrations and resonances generated in the epilarynx and

pharynx. It focuses on the long-term auditory-articulatory component of

accent in the languages of the world, explaining how voice quality relates to

segmental and syllabic sounds. Phonetic illustrations of phonation types

and of laryngeal and oral vocal tract articulatory postures are provided.

Extensive video and audio material is available on a companion website.

The book presents computational simulations, the laryngeal and voice

quality foundations of infant speech acquisition, speech/voice disorders and

surgeries that entail compensatory laryngeal articulator adjustment, and an

exploration of the role of voice quality in sound change and of the larynx in

the evolution of speech.


1. Voice and voice quality; 2. Voice quality classification; 3. Instrumental case

studies and computational simulations of voice quality; 4. Linguistic, paralinguistic

and extralinguistic illustrations of voice quality; 5. Phonological implications of

voice quality theory; 6. Infant acquisition of speech and voice quality; 7. Clinical

illustrations of voice quality; 8. Laryngeal articulation and voice quality in sound

change, language ontogeny.


5-1-6Albert di Cristo,' Les langues naturelles'. HAL archive ouverte

Albert di Cristo, les langues naturelles.
Première partie : La structure informationnelle et ses déterminants
, 2019, 548 p.

Cet ouvrage constitue la première partie d?un vaste travail dédié à l?étude des façons dont les langues naturelles conditionnent l?information et au rôle que joue la prosodie dans l?expression de ces conditionnements. Cette première partie se propose d?analyser, sous ses divers aspects (principalement d'ordre épistémologiques), la notion de structure informationnelle, notamment dans ses relations avec la grammaire et d?examiner, dans le détail, les déterminants qui forment l?armature de cette structure. Dans cette perspective, les discussions portent, outre sur les notions de thème, de topique et de « given », sur celles de focus, de focalisation et de contraste, qui font l?objet d?analyses approfondies. Les discussions s?attachent à appréhender ces notions, à la fois dans l?optique de leurs propriétés formelles, de leur fonctionnalité et des significations qu?elles contribuent à délivrer. Un chapitre entier de cette première partie est consacré à l?étude du questionnement et à la manière dont l?organisation de l?information est gérée dans l?exercice de cette activité. L?ouvrage est doté d?une bibliographie qui comporte plus de deux mille références.

Cet ouvrage sera complété par une 2ème partie, en cours de rédaction, qui traitera essentiellement de la prosodie et de son rôle dans les conditionnements de l'information.


5-1-7Benjamin Weiss, 'Talker Quality in Human and Machine Interaction - Modeling the Listener’s Perspective in Passive and Interactive Scenarios'. T-Labs Series in Telecommunication Services. Springer Nature, Cham. (2020)

Benjamin Weiss (2020): 'Talker Quality in Human and Machine Interaction - Modeling the Listener’s Perspective in Passive and Interactive Scenarios'. T-Labs Series in Telecommunication Services. Springer Nature, Cham.

In this book, the background, state of research, and own contributions to the assessment and prediction of talker quality that is constituted in voice perception and in dialog are presented. Starting from theories and empirical findings from human interaction, major results and approaches are transferred to the domain of human-computer interaction. The main subject of this book is to contribute to the evaluation of spoken interaction in both humans and between human and computer, and in particular to the quality subsequently attributed to the speaking system or person, based on the listening and interactive experience.


5-1-8W.F.Katz, P.F.Assman, 'The Routledge Handbook of Phonetics', Routledge.


The Routledge Handbook of Phonetics Edited by William F. Katz and Peter F. Assmann The Routledge Handbook of Phonetics provides a comprehensive and up-to-date compilation of research, history and techniques in phonetics. With contributions from 41 prominent authors from North America, Europe, Australia and Japan, and including over 130 figures to illustrate key points, this handbook covers all the most important areas in the field, including:

  • The history and scope of techniques used, including speech synthesis, vocal tract imaging techniques, and obtaining information on under-researched languages from language archives;
  • The physiological bases of speech and hearing, including auditory, articulatory, and neural explanations of hearing, speech, and language processes;
  • Theories and models of speech perception and production related to the processing of consonants, vowels, prosody, tone, and intonation;



5-1-9Proceedings of SLTU-CCURL2020

Dear all,

we are very happy to announce that the SLTU-CCURL2020 Proceedings are available online:
This year, LREC2020 would have featured an extraordinary event: the first joint SLTU-CCURL2020 Workshop, which was planned as a two-day workshop, with 54 papers accepted either as oral and poster presentations.
The workshop program was enriched by two tutorials and two keynote speeches.
We will miss the presentations, the discussions and the overall stimulating environment very deeply. 
We are thankful to ELRA and ISCA for their support to the workshop, to our Google sponsor and to the 60 experts of the Program Committee, who worked tirelessly in order to help us to select the best papers representing a wide perspective over NLP, speech and computational linguistics addressing less-resource languages.
Looking forward to better times when we will be able to meet in person again, we hope that you will find these workshop proceedings relevant and stimulating for your own research.
With our best wishes,

Claudia Soria, Laurent Besacier, Dorothee Beermann, and Sakriani Sakti

5-2 Database
5-2-1Linguistic Data Consortium (LDC) update (August 2020)


In this newsletter:
LDC adds DOI Identifier to its Language Resources
Fall 2020 LDC Data Scholarship Program

New Publications:
LORELEI Vietnamese Representative Language Pack

DEFT Chinese Light and Rich ERE Annotation
CALLFRIEND American English – Southern Dialect Second Edition





LDC adds DOI Identifier to its Language Resources
As of July 2020, LDC’s language resources include a Digital Object Identifier (DOI), an internationally recognized identification standard for online digital material. DOIs are alpha numeric strings that correspond to URLs and metadata for specified resources. They are expressed as links that resolve to the object’s online location. For example, the DOI for Penn Parsed Corpora of Historical English LDC2020T16 is, which leads users to the LDC catalog entry for this data set. To facilitate its assignment and administration of DOIs, LDC has joined DataCite, a global DOI provider for research data. (DOIs for resources released before July 2020 will be assigned through a process expected to be completed shortly.) LDC data sets now have four persistent identifiers: a unique LDC number, ISBN, ISLRN, and DOI. Adding DOIs is consistent with our aim to follow best practices for archiving and curating digital resources, evidenced by the CoreTrustSeal certification which recognizes the LDC Catalog as a trustworthy data repository.

Fall 2020 LDC Data Scholarship Program
Student applications for the Fall 2020 LDC Data Scholarship program are being accepted now through September 15, 2020. This scholarship program provides eligible students with no-cost access to LDC data. Students must complete an application consisting of a data use proposal and letter of support from their advisor.

For application requirements and program rules, visit the LDC Data Scholarship page.







New publications:
(1) LORELEI Vietnamese Representative Language Pack consists of Vietnamese monolingual text, Vietnamese-English parallel text, annotations, supplemental resources, and related software tools developed by LDC for the DARPA LORELEI program.

The LORELEI (Low Resource Languages for Emergent Incidents) program was concerned with building human language technology for low resource languages in the context of emergent situations like natural disasters or disease outbreaks. Linguistic resources for LORELEI include Representative Language Packs and Incident Language Packs for over two dozen low resource languages, comprising data, annotations, basic natural language processing tools, lexicons, and grammatical resources. Representative languages were selected to provide broad typological coverage, while incident languages were selected to evaluate system performance on a language whose identity was disclosed at the start of the evaluation.

Data was collected in the following genres: discussion forum, news, reference, social network, and weblogs. Data volumes are as follows:


  • Over 172 million words of Vietnamese monolingual text, approximately 325,000 words of which were translated into English
  • 106,000 Vietnamese words translated from English data
  • 1.9 million words of found parallel text


Approximately 75,000 words were annotated for named entities and up to 25,000 words contain additional annotation, including situation frames (identifying entities, needs, and issues) and entity linking and detection.

LORELEI Vietnamese Representative Language Pack is distributed via web download.

2020 Subscription Members will automatically receive copies of this corpus. 2020 Standard Members may request a copy as part of their 16 free membership corpora. Non-members may license this data for a $250.


(2) DEFT Chinese Light and Rich ERE Annotation contains Chinese discussion forum web text annotated for entities, relations, and events (ERE) using the ERE Light and ERE Rich annotations schemas developed by LDC. Light ERE annotation labels entity mentions for the target set of ERE types between and among those entities, including coreference. Rich ERE annotation expands types and tagging for ERE annotation tasks and replaces event coreference with event hopper annotation. All files in this release (157) were annotated following Light ERE guidelines; a subset (149) were also labeled with Rich ERE annotation. 

DARPA’s Deep Exploration and Filtering of Text (DEFT) program aimed to address remaining capability gaps in state-of-the-art natural language processing technologies related to inference, causal relationships, and anomaly detection. LDC supported the DEFT program by collecting, creating, and annotating a variety of data sources.

DEFT Chinese Light and Rich ERE Annotation is distributed via web download.

2020 Subscription Members will automatically receive copies of this corpus. 2020 Standard Members may request a copy as part of their 16 free membership corpora. Non-members may license this data for $2000.


(3) CALLFRIEND American English – Southern Dialect Second Edition was developed by LDC and consists of approximately 26 hours of unscripted telephone conversations between native speakers of Southern dialects of American English. This second edition updates the audio files to wav format, simplifies the directory structure, and adds documentation and metadata. The first edition is available as CALLFRIEND American English-Southern Dialect (LDC96S47).

The CALLFRIEND collection was conducted by LDC in support of language identification technology development. All data in this release was collected before July 1997. Participants could speak with a person of their choice on any topic; most called family members and friends. All calls originated in North America. The recorded conversations last up to 30 minutes.

CALLFRIEND American English – Southern Dialect Second Edition is distributed via web download.

2020 Subscription Members will automatically receive copies of this corpus. 2020 Standard Members may request a copy as part of their 16 free membership corpora. Non-members may license this data for $1000.






5-2-2ELRA - Language Resources Catalogue - Update (May 2020)
We are happy to announce that 1 new Speech resource is now available in our catalogue.
Mandarin Chinese Conversational Recognition Corpus from Magic Data Tech
This dataset consists of 4.98 hours of transcribed conversational speech in Mandarin Chinese, where 30 conversations are uttered by 32 speakers (16 males and 16 females). The audios are sampled at 16 kHz and quantized at 16 bits.
For each conversation, there are two close-talking channels recorded via the microphones, one for each speaker, as well as three far-field channels recorded by iPhone, Androïd Phone, and recorder respectively.

This corpus may be obtained as a complete set or by selecting specific channels (two close-talking channels shall be understood as 1 single channel):

ELRA-S0409-01 MDT Mandarin Chinese Conversational Recognition Corpus - complete set
ISLRN: 559-956-475-937-1
For more information, see:
ELRA-S0409-02 MDT Mandarin Chinese Conversational Recognition Corpus - 1 channel
ISLRN: 234-140-315-272-4
For more information, see:
ELRA-S0409-03 MDT Mandarin Chinese Conversational Recognition Corpus - 2 channels
ISLRN: 383-054-806-637-3
For more information, see:
ELRA-S0409-04 MDT Mandarin Chinese Conversational Recognition Corpus - 3 channels
ISLRN: 235-882-638-211-2
For more information, see:

For more information on the catalogue, please contact Valérie Mapelli

If you would like to enquire about having your resources distributed by ELRA, please do not hesitate to contact us.

Visit our On-line Catalogue:
Visit the Universal Catalogue:
Archives of ELRA Language Resources Catalogue Updates:



5-2-3Speechocean – update (August 2019)


English Speech Recognition Corpus - Speechocean


At present, Speechocean has produced more than 24,000 hours of English Speech Recognition Corpora, including some rare corpora recorded by kids. Those corpora were recorded by 23,000 speakers in total. Please check the form below:





American English



Indian English



British English



Australian English



Chinese (Mainland) English



Canadian English



Japanese English



Singapore English



Russian English



Romanian English



French English



Chinese (Hong Kong) English



Italian English



Portugal English



Spainish English



German English



Korean English



Indonesian English





If you have any further inquiries, please do not hesitate to contact us.













5-2-4Google 's Language Model benchmark
 Here is a brief description of the project.

'The purpose of the project is to make available a standard training and test setup for language modeling experiments.

The training/held-out data was produced from a download at using a combination of Bash shell and Perl scripts distributed here.

This also means that your results on this data set are reproducible by the research community at large.

Besides the scripts needed to rebuild the training/held-out data, it also makes available log-probability values for each word in each of ten held-out data sets, for each of the following baseline models:

  • unpruned Katz (1.1B n-grams),
  • pruned Katz (~15M n-grams),
  • unpruned Interpolated Kneser-Ney (1.1B n-grams),
  • pruned Interpolated Kneser-Ney (~15M n-grams)


Happy benchmarking!'


5-2-5Forensic database of voice recordings of 500+ Australian English speakers

Forensic database of voice recordings of 500+ Australian English speakers

We are pleased to announce that the forensic database of voice recordings of 500+ Australian English speakers is now published.

The database was collected by the Forensic Voice Comparison Laboratory, School of Electrical Engineering & Telecommunications, University of New South Wales as part of the Australian Research Council funded Linkage Project on making demonstrably valid and reliable forensic voice comparison a practical everyday reality in Australia. The project was conducted in partnership with: Australian Federal Police,  New South Wales Police,  Queensland Police, National Institute of Forensic Sciences, Australasian Speech Sciences and Technology Association, Guardia Civil, Universidad Autónoma de Madrid.

The database includes multiple non-contemporaneous recordings of most speakers. Each speaker is recorded in three different speaking styles representative of some common styles found in forensic casework. Recordings are recorded under high-quality conditions and extraneous noises and crosstalk have been manually removed. The high-quality audio can be processed to reflect recording conditions found in forensic casework.

The database can be accessed at:


5-2-6Audio and Electroglottographic speech recordings


Audio and Electroglottographic speech recordings from several languages

We are happy to announce the public availability of speech recordings made as part of the UCLA project 'Production and Perception of Linguistic Voice Quality'.

Audio and EGG recordings are available for Bo, Gujarati, Hmong, Mandarin, Black Miao, Southern Yi, Santiago Matatlan/ San Juan Guelavia Zapotec; audio recordings (no EGG) are available for English and Mandarin. Recordings of Jalapa Mazatec extracted from the UCLA Phonetic Archive are also posted. All recordings are accompanied by explanatory notes and wordlists, and most are accompanied by Praat textgrids that locate target segments of interest to our project.

Analysis software developed as part of the project – VoiceSauce for audio analysis and EggWorks for EGG analysis – and all project publications are also available from this site. All preliminary analyses of the recordings using these tools (i.e. acoustic and EGG parameter values extracted from the recordings) are posted on the site in large data spreadsheets.

All of these materials are made freely available under a Creative Commons Attribution-NonCommercial-ShareAlike-3.0 Unported License.

This project was funded by NSF grant BCS-0720304 to Pat Keating, Abeer Alwan and Jody Kreiman of UCLA, and Christina Esposito of Macalester College.

Pat Keating (UCLA)


5-2-7EEG-face tracking- audio 24 GB data set Kara One, Toronto, Canada

We are making 24 GB of a new dataset, called Kara One, freely available. This database combines 3 modalities (EEG, face tracking, and audio) during imagined and articulated speech using phonologically-relevant phonemic and single-word prompts. It is the result of a collaboration between the Toronto Rehabilitation Institute (in the University Health Network) and the Department of Computer Science at the University of Toronto.


In the associated paper (abstract below), we show how to accurately classify imagined phonological categories solely from EEG data. Specifically, we obtain up to 90% accuracy in classifying imagined consonants from imagined vowels and up to 95% accuracy in classifying stimulus from active imagination states using advanced deep-belief networks.


Data from 14 participants are available here:


If you have any questions, please contact Frank Rudzicz at


Best regards,




PAPER Shunan Zhao and Frank Rudzicz (2015) Classifying phonological categories in imagined and articulated speech. In Proceedings of ICASSP 2015, Brisbane Australia

ABSTRACT This paper presents a new dataset combining 3 modalities (EEG, facial, and audio) during imagined and vocalized phonemic and single-word prompts. We pre-process the EEG data, compute features for all 3 modalities, and perform binary classi?cation of phonological categories using a combination of these modalities. For example, a deep-belief network obtains accuracies over 90% on identifying consonants, which is signi?cantly more accurate than two baseline supportvectormachines. Wealsoclassifybetweenthedifferent states (resting, stimuli, active thinking) of the recording, achievingaccuraciesof95%. Thesedatamaybeusedtolearn multimodal relationships, and to develop silent-speech and brain-computer interfaces.



5-2-8TORGO data base free for academic use.

In the spirit of the season, I would like to announce the immediate availability of the TORGO database free, in perpetuity for academic use. This database combines acoustics and electromagnetic articulography from 8 individuals with speech disorders and 7 without, and totals over 18 GB. These data can be used for multimodal models (e.g., for acoustic-articulatory inversion), models of pathology, and augmented speech recognition, for example. More information (and the database itself) can be found here:



Datatang is a global leading data provider that specialized in data customized solution, focusing in variety speech, image, and text data collection, annotation, crowdsourcing services.


Summary of the new datasets (2018) and a brief plan for 2019.




? Speech data (with annotation) that we finished in 2018 


Datasets Length
  ( Hours )
British English
Spanish (Mexico/Colombia)
Brazilian Portuguese
European Portuguese


?2019 ongoing  speech project 



Project Name

Europeans speak English

1000 Hours-Spanish Speak English

1000 Hours-French Speak English

1000 Hours-German Speak English

Call Center Speech

1000 Hours-Call Center Speech

off-the-shelf data expansion

1000 Hours-Chinese Speak English

1500 Hours-Mixed Chinese and English Speech Data




On top of the above,  there are more planed speech data collections, such as Japanese speech data, children`s speech data, dialect speech data and so on.  


What is more, we will continually provide those data at a competitive price with a maintained high accuracy rate.




If you have any questions or need more details, do not hesitate to contact us 


It would be possible to send you with a sample or specification of the data.





5-2-10Fearless Steps Corpus (University of Texas, Dallas)

Fearless Steps Corpus

John H.L. Hansen, Abhijeet Sangwan, Lakshmish Kaushik, Chengzhu Yu Center for Robust Speech Systems (CRSS), Eric Jonsson School of Engineering, The University of Texas at Dallas (UTD), Richardson, Texas, U.S.A.

NASA’s Apollo program is a great achievement of mankind in the 20th century. CRSS, UT-Dallas has undertaken an enormous Apollo data digitization initiative where we proposed to digitize Apollo mission speech data (~100,000 hours) and develop Spoken Language Technology based algorithms to analyze and understand various aspects of conversational speech. Towards achieving this goal, a new 30 track analog audio decoder is designed to decode 30 track Apollo analog tapes and is mounted on to the NASA Soundscriber analog audio decoder (in place of single channel decoder). Using the new decoder all 30 channels of data can be decoded simultaneously thereby reducing the digitization time significantly. 
We have digitized 19,000 hours of data from Apollo missions (including entire Apollo-11, most of Apollo-13, Apollo-1, and Gemini-8 missions). This audio archive is named as “Fearless Steps Corpus”. This is one of the most unique and singularly large naturalistic audio corpus of such magnitude. Automated transcripts are generated by building Apollo mission specific custom Deep Neural Networks (DNN) based Automatic Speech Recognition (ASR) system along with Apollo mission specific language models. Speaker Identification System (SID) to identify the speakers are designed. A complete diarization pipeline is established to study and develop various SLT tasks. 
We will release this corpus for public usage as a part of public outreach and promote SLT community to utilize this opportunity to build naturalistic spoken language technology systems. The data provides ample opportunity setup challenging tasks in various SLT areas. As a part of this outreach we will be setting “Fearless Challenge” in the upcoming INTERSPEECH 2018. We will define and propose 5 tasks as a part of this challenge. The guidelines and challenge data will be released in the Spring 2018 and will be available for download for free. The five challenges are, (1) Automatic Speech Recognition (2) Speaker Identification (3) Speech Activity Detection (4) Speaker Diarization (5) Keyword spotting and Joint Topic/Sentiment detection.
Looking forward for your participation ( 


5-2-11SIWIS French Speech Synthesis Database
The SIWIS French Speech Synthesis Database includes high quality French speech recordings and associated text files, aimed at building TTS systems, investigate multiple styles, and emphasis. A total of 9750 utterances from various sources such as parliament debates and novels were uttered by a professional French voice talent. A subset of the database contains emphasised words in many different contexts. The database includes more than ten hours of speech data and is freely available.

5-2-12JLCorpus - Emotional Speech corpus with primary and secondary emotions
JLCorpus - Emotional Speech corpus with primary and secondary emotions:

For further understanding the wide array of emotions embedded in human speech, we are introducing an emotional speech corpus. In contrast to the existing speech corpora, this corpus was constructed by maintaining an equal distribution of 4 long vowels in New Zealand English. This balance is to facilitate emotion related formant and glottal source feature comparison studies. Also, the corpus has 5 secondary emotions along with 5 primary emotions. Secondary emotions are important in Human-Robot Interaction (HRI), where the aim is to model natural conversations among humans and robots. But there are very few existing speech resources to study these emotions,and this work adds a speech corpus containing some secondary emotions.

Please use the corpus for emotional speech related studies. When you use it please include the citation as:

Jesin James, Li Tian, Catherine Watson, 'An Open Source Emotional Speech Corpus for Human Robot Interaction Applications', in Proc. Interspeech, 2018.

To access the whole corpus including the recording supporting files, click the following link:, (if you have already installed the Kaggle API, you can type the following command to download: kaggle datasets download -d tli725/jl-corpus)

Or if you simply want the raw audio+txt files, click the following link:

The corpus was evaluated by a large scale human perception test with 120 participants. The link to the survey are here- For Primary emorion corpus:

For Secondary emotion corpus:

These surveys will give an overall idea about the type of recordings in the corpus.

The perceptually verified and annotated JL corpus will be given public access soon.


5-2-13OPENGLOT –An open environment for the evaluation of glottal inverse filtering

OPENGLOT –An open environment for the evaluation of glottal inverse filtering


OPENGLOT is a publically available database that was designed primarily for the evaluation of glottal inverse filtering algorithms. In addition, the database can be used in evaluating formant estimation methods. OPENGLOT consists of four repositories. Repository I contains synthetic glottal flow waveforms, and speech signals generated by using the Liljencrants–Fant (LF) waveform as an excitation, and an all-pole vocal tract model. Repository II contains glottal flow and speech pressure signals generated using physical modelling of human speech production. Repository III contains pairs of glottal excitation and speech pressure signal generated by exciting 3D printed plastic vocal tract replica with LF excitations via a loudspeaker. Finally, Repository IV contains multichannel recordings (speech pressure signal, EGG, high-speed video of the vocal folds) from natural production of speech.


OPENGLOT is available at:


5-2-14Corpus Rhapsodie

Nous sommes heureux de vous annoncer la publication d¹un ouvrage consacré
au treebank Rhapsodie, un corpus de français parlé de 33 000 mots
finement annoté en prosodie et en syntaxe.

Accès à la publication : (voir flyer

Accès au treebank :
Les données librement accessibles sont diffusées sous licence Creative
Le site donne également accès aux guides d¹annotations.


5-2-15The My Science Tutor Children?s Conversational Speech Corpus (MyST Corpus) , Boulder Learning Inc.

The My Science Tutor Children?s Conversational Speech Corpus (MyST Corpus) is the world?s largest English children?s speech corpus.  It is freely available to the research community for research use.  Companies can acquire the corpus for $10,000.  The MyST Corpus was collected over a 10-year period, with support from over $9 million in grants from the US National Science Foundation and Department of Education, awarded to Boulder Learning Inc. (Wayne Ward, Principal Investigator).

The MyST corpus contains speech collected from 1,374 third, fourth and fifth grade students.  The students engaged in spoken dialogs with a virtual science tutor in 8 areas of science.  A total of 11,398 student sessions of 15 to 20 minutes produced a total of 244,069 utterances.  42% of the utterances have been transcribed at the word level.  The corpus is partitioned into training and test sets to support comparison of research results across labs. All parents and students signed consent forms, approved by the University of Colorado?s Institutional Review Board,  that authorize distribution of the corpus for research and commercial use. 

The MyST children?s speech corpus contains approximately ten times as many spoken utterances as all other English children?s speech corpora combined (see 

Additional information about the corpus, and instructions for how to acquire the corpus (and samples of the speech data) can be found on the Boulder Learning Web site at   


5-2-16HARVARD speech corpus - native British English speaker
  • HARVARD speech corpus - native British English speaker, digital re-recording

5-2-17Magic Data Technology Kid Voice TTS Corpus in Mandarin Chinese (November 2019)

Magic Data Technology Kid Voice TTS Corpus in Mandarin Chinese


Magic Data Technology is one of the leading artificial intelligence data service providers in the world. The company is committed to providing a wild range of customized data services in the fields of speech recognition, intelligent imaging and Natural Language Understanding.


This corpus was recorded by a four-year-old Chinese girl originally born in Beijing China. This time we published 15-minute speech data from the corpus for non-commercial use.


The contents and the corresponding descriptions of the corpus:

  • The corpus contains 15 minutes of speech data, which is recorded in NC-20 acoustic studio.

  • The speaker is 4 years old originally born in Beijing

  • Detail information such as speech data coding and speaker information is preserved in the metadata file.

  • This corpus is natural kid style.

  • Annotation includes four parts: pronunciation proofreading, prosody labeling, phone boundary labeling and POS Tagging.

  • The annotation accuracy is higher than 99%.

  • For phone labeling, the database contains the annotation not only on the boundary of phonemes, but also on the boundary of the silence parts.


The corpus aims to help researchers in the TTS fields. And it is part of a much bigger dataset (2.3 hours MAGICDATA Kid Voice TTS Corpus in Mandarin Chinese) which was recorded in the same environment. This is the first time to publish this voice!


Please note that this corpus has got the speaker and her parents’ authorization.


Samples are available.

Do not hesitate to contact us for any questions.




5-2-18FlauBERT: a French LM
Here is FlauBERT: a French LM learnt (with #CNRS J-Zay supercomputer) on a large and heterogeneous corpus. Along with it comes FLUE (evaluation setup for French NLP). FlauBERT was successfully applied to complex tasks (NLI, WSD, Parsing).  More on
More details on this online paper: 

5-2-19ELRA-S0408 SpeechTera Pronunciation Dictionary

ELRA-S0408 Speechtera Pronunciation Dictionary

ISLRN: 645-563-102-594-8
The SpeechTera Pronunciation Dictionary is a machine-readable pronunciation dictionary for Brazilian Portuguese and comprises 737,347 entries. Its phonetic transcription is based on 13 linguistics varieties spoken in Brazil and contains the pronunciation of the frequent word forms found in the transcription data of the SpeechTera's speech and text database (literary, newspaper, movies, miscellaneous). Each one of the thirteen dialects comprises 56,719 entries.
For more information, see:

For more information on the catalogue, please contact Valérie Mapelli

If you would like to enquire about having your resources distributed by ELRA, please do not hesitate to contact us.

Visit our On-line Catalogue:
Visit the Universal Catalogue:
Archives of ELRA Language Resources Catalogue Updates:


5-2-20Ressources of ELRC Network

Paris, France, April 23, 2020

ELRA is happy to announce that Language Resources collected within the ELRC Network, funded by the European Commission, are now available from the ELRA Catalogue of Language Resources.

In total, 180 Written Corpora, 5 Multilingual Lexicons and 2 Terminological Resources, are freely available under open licences and can be downloaded directly from the catalogue. Type 'ELRC' in the catalogue search engine ( to access and download resources.

All these Language Resources can be used to support your Machine Translation solutions developments. They cover the official languages of the European Union and CEF associated countries.

More LRs coming from ELRC will be added as they become available.

About ELRC
ELRC (European Language Resources Coordination) Network raises awareness and promote the acquisition and continued identification and collection of language resources in all official languages of the EU and CEF associated countries. These activities aim to help to improve the quality, coverage and performance of automated translation solutions in the context of current and future CEF digital services.

To find out more about ELRC, please visit the website:

About ELRA
The European Language Resources Association (ELRA) is a non-profit-making organisation founded by the European Commission in 1995, with the mission of providing a clearing house for Language Resources and promoting Human Language Technologies. Language Resources covering various fields of HLT (including Multimodal, Speech, Written, Terminology) and a great number of languages are available from the ELRA catalogue. ELRA's strong involvement in the fields of Language Resources  and Language Technologies is also emphasized at the LREC conference, organized every other year since 1998.

To find out more about ELRA, please visit the website:

For more information on the catalogue, please contact Valérie Mapelli
If you would like to enquire about having your resources distributed by ELRA, please do not hesitate to contact us.

Visit our On-line Catalogue:
Visit the Universal Catalogue:
Archives of ELRA Language Resources Catalogue Updates:


5-2-21Language Resources distribution agreement between ELRA and SpeechOcean

Press Release - Immediate
Paris, France, May 4, 2020

ELRA and SpeechOcean signed a new Language Resources distribution agreement. On behalf of ELRA, ELDA acts as the distribution agency for SpeechOcean since 2007 and incorporated 46 new speech resources to the ELRA Catalogue of Language Resources catalogue.

Those resources were designed and collected to boost Speech Recognition. They cover the following languages:

  • Chinese-Mandarin,
  • French from France and Canada,
  • German,
  • Italian,
  • Hong Kong Cantonese,
  • Japanese,
  • Korean,
  • Pashto,
  • Portuguese from Portugal and Brazil,
  • Russian,
  • Spanish from US and Mexico,
  • Taiwanese,
  • Several variants of English (English from Australia, Canada, China, Japan, Korea, United Kingdom, USA)

This new set of speech LRs leads to a total of 103 LRs distributed by ELRA on behalf of SpeechOcean, and aims to strengthen ELRA's position as the leading worldwide distribution centre. With this agreement SpeechOcean will get more visibility in particular on the European market.

The detailed list of all 103 Language Resources from SpeechOcean is available here:

About SpeechOcean
Since its establishment as an AI data resource provider, Speechocean has always devoted itself to providing specialized engineering data products and services to enterprises and scientific research institutions in the whole industry chain of AI. Our business involves various domains such as speech recognition, speech synthesis, computer vision, lexicon, and natural language processing and provides relevant services for the design, collection, transcription, annotation, etc. of data.

To find out more about SpeechOcean, please visit the website:

About ELRA
The European Language Resources Association (ELRA) is a non-profit-making organisation founded by the European Commission in 1995, with the mission of providing a clearing house for Language Resources and promoting Human Language Technologies. Language Resources covering various fields of HLT (including Multimodal, Speech, Written, Terminology) and a great number of languages are available from the ELRA catalogue. ELRA's strong involvement in the fields of Language Resources  and Language Technologies is also emphasized at the LREC conference, organized every other year since 1998.

To find out more about ELRA, please visit the website:

For more information on the catalogue, please contact Valérie Mapelli

If you would like to enquire about having your resources distributed by ELRA, please do not hesitate to contact us.

Visit our On-line Catalogue:
Visit the Universal Catalogue:
Archives of ELRA Language Resources Catalogue Updates:


5-2-22Sharing Language Ressourses via ELRA

ELRA recognises the importance of sharing Language Resources (LRs) and making them available to the community.

Since the 2014 edition of LREC, the Language Resources and Evaluation Conference, participants have been offered the possibility to share their LRs (data, tools, web-services, etc.) when submitting a paper, uploading them in a special LREC repository set up by ELRA. This effort of sharing LRs, linked to the LRE Map initiative ( for their description, contributes to creating a common repository where everyone can deposit and share data.

The LREC initiative 'Share your LRs' was launched in 2014 in Reykjavik. It was successfully continued in 2016 in Portoro? and 2018 in Miyazaki.

Corresponding repositories are available here:

For more information and/or questions, please write to

5-2-23ELRA announces that MEDIA data are now available for free for academic research

ELRA announces that MEDIA data are now available for free for academic research

Further to the request of the HLT French community to foster evaluation activities for man-machine dialogue systems for French language, ELRA has decided to provide a free access to the MEDIA speech corpora and evaluation package for academic research purposes.

The MEDIA data can be found in the ELRA Catalogue under the following references:

Data available from the ELRA Catalogue can be obtained easily by contacting ELRA.  

The MEDIA project was carried out within the framework of Technolangue, the French national research programme funded by the French Ministry of Research and New Technologies (MRNT) with the objective of running a campaign for the evaluation of man-machine dialogue systems for French. The campaign was distributed over two actions: an evaluation taking into account the dialogue context and an evaluation not taking into account the dialogue context.

PortMedia was a follow up project supported by the French Research Agency (ANR). The French and Italian corpus was produced by ELDA, with the same paradigm and specifications as the MEDIA speech database but on a different domain.

For more information and/or questions, please write to

 *** About ELRA ***
The European Language Resources Association (ELRA) is a non-profit making organisation founded by the European Commission in 1995, with the mission of providing a clearing house for language resources and promoting Human Language Technologies (HLT).

To find out more about ELRA and its respective catalogue, please visit: and


5-3 Software
5-3-1Cantor Digitalis, an open-source real-time singing synthesizer controlled by hand gestures.

We are glad to announce the public realease of the Cantor Digitalis, an open-source real-time singing synthesizer controlled by hand gestures.

It can be used e.g. for making music or for singing voice pedagogy.

A wide variety of voices are available, from the classic vocal quartet (soprano, alto, tenor, bass), to the extreme colors of childish, breathy, roaring, etc. voices.  All the features of vocal sounds are entirely under control, as the synthesis method is based on a mathematic model of voice production, without prerecording segments.

The instrument is controlled using chironomy, i.e. hand gestures, with the help of interfaces like stylus or fingers on a graphic tablet, or computer mouse. Vocal dimensions such as the melody, vocal effort, vowel, voice tension, vocal tract size, breathiness etc. can easily and continuously be controlled during performance, and special voices can be prepared in advance or using presets.

Check out the capabilities of Cantor Digitalis, through performances extracts from the ensemble Chorus Digitalis:

In pratice, this release provides:
  • the synthesizer application
  • the source code in the form of a Max package (GPL-like license)
  • a documentation for the musician and another for the developper
What do you need ?
  • a Mac OSX
  • ideally a Wacom graphic tablet, but it also works with your computer mouse
  • for the developers, the Max software
Interested ?
  • To download the Cantor Digitalis, click here
  • To subscribe to the Cantor Digitalisnewsletter and/or the forum list, or to contact the developers, click here
  • To learn about the Chorus Digitalis, ensemble of Cantor Digitalisand watch videos of performances, click here
  • For more details about the Cantor Digitalis, click here
The Cantor Digitalis team (who loves feedback —
Christophe d'Alessandro, Lionel Feugère, Olivier Perrotin

5-3-2MultiVec: a Multilingual and MultiLevel Representation Learning Toolkit for NLP


We are happy to announce the release of our new toolkit “MultiVec” for computing continuous representations for text at different granularity levels (word-level or sequences of words). MultiVec includes Mikolov et al. [2013b]’s word2vec features, Le and Mikolov [2014]’s paragraph vector (batch and online) and Luong et al. [2015]’s model for bilingual distributed representations. MultiVec also includes different distance measures between words and sequences of words. The toolkit is written in C++ and is aimed at being fast (in the same order of magnitude as word2vec), easy to use, and easy to extend. It has been evaluated on several NLP tasks: the analogical reasoning task, sentiment analysis, and crosslingual document classification. The toolkit also includes C++ and Python libraries, that you can use to query bilingual and monolingual models.


The project is fully open to future contributions. The code is provided on the project webpage ( with installation instructions and command-line usage examples.


When you use this toolkit, please cite:



Title                    = {{MultiVec: a Multilingual and MultiLevel Representation Learning Toolkit for NLP}},

Author                   = {Alexandre Bérard and Christophe Servan and Olivier Pietquin and Laurent Besacier},

Booktitle                = {The 10th edition of the Language Resources and Evaluation Conference (LREC 2016)},

Year                     = {2016},

Month                    = {May}



The paper is available here:


Best regards,


Alexandre Bérard, Christophe Servan, Olivier Pietquin and Laurent Besacier


5-3-3An android application for speech data collection LIG_AIKUMA
We are pleased to announce the release of LIG_AIKUMA, an android application for speech data collection, specially dedicated to language documentation. LIG_AIKUMA is an improved version of the Android application (AIKUMA) initially developed by Steven Bird and colleagues. Features were added to the app in order to facilitate the collection of parallel speech data in line with the requirements of a French-German project (ANR/DFG BULB - Breaking the Unwritten Language Barrier). 
The resulting app, called LIG-AIKUMA, runs on various mobile phones and tablets and proposes a range of different speech collection modes (recording, respeaking, translation and elicitation). It was used for field data collections in Congo-Brazzaville resulting in a total of over 80 hours of speech.
Users who just want to use the app without access to the code can download it directly from the forge direct link: 
Code is also available on demand (contact and
More details on LIG_AIKUMA can be found on the following paper:

5-3-4Web services via ALL GO from IRISA-CNRS

It is our pleasure to introduce A||GO ( or, a platform providing a collection of web-services for the automatic analysis of various data, including multimedia content across modalities. The platform builds on the back-end web service deployment infrastructure developed and maintained by  Inria?s  Service for Experimentation and Development (SED). Originally dedicated to multimedia content, A||GO progressively broadened to other fields such as computational biology, networks and telecommunications, computational graphics or computational physics.

As part of the CNRS PlaSciDo initiative [1], the Linkmedia team at IRISA / Inria Rennes is making available via A||GO a number of web services devoted to multimedia content analysis across modalities (language, audio, image, video). The web services provided currently include research results from the Linkmedia team as well as contribution from a number of partners. A list of the services available by the date is given below and the current state is available at along with demo videos. Most web services are interoperable, facilitating the implementation of a multimedia content analysis processing chain, and are free to use for trial, prototyping or lab work. A brief and free account creation step will allow you to execute the web-services using either the graphical interface or a command line via a dedicated API.

We expect the number of web services to grow over time and invite interested parties to contact us should they wish to contribute the multimedia web service offer of A||GO.

List of multimedia content analysis tools currently available on A||GO:
- Audio Processing
        SaMuSa: music/speech segmentation
        SilAD: silence detection repeated audio motif discovery
        LORIA STS v2: speech transcription for the French language from LORIA
        Multi channel BSS locate: audio source localization toolbox from IRISA-PANAMA
        A-spade: audio declipper from IRISA-PANAMA
        Transvox: voice faker from LORIA
- Natural Language Processing
        NERO: name entity recognition
        TermEx: keywords/indexing terms detection
        Otis!: topic segmentation
        Hi-tost: hierarchical topic structuring
- Video Processing
        Vidseg: video shot segmentation
        HUFA: face detection and tracking
Shortcuts to Linkmedia services are also available here:
For more information don't hesitate to contact us (

Gabriel Sargent and Guillaume Gravier
Rennes, France


5-3-5Clickable map - Illustrations of the IPA

Clickable map - Illustrations of the IPA

We have produced a clickable map showing the Illustrations of the International Phonetic

The map is being updated with each new issue of the Journal of the International Phonetic

Marija Tabain - La Trobe University, Australia
Richard Beare - Monash University & MCRI, Australia


5-3-6LIG-Aikuma running on mobile phones and tablets


Dear all,

LIG is pleased to inform you that the website for the app Lig-Aikuma is online:
In the same time, an update of Lig-Aikuma (V3) was made available (see website).      

LIG-AIKUMA is a free Android app running on various mobile phones and tablets. The app proposes a range of different speech collection modes (recording, respeaking, translation and elicitation) and offers the possibility to share recordings between users. LIG-AIKUMA is built upon the initial AIKUMA app developed by S. Bird & F. Hanke (see  for more information)

Improvements of the app:

  • Visual upgrade:
    + Waveform visualizer on the Respeaking and Translation modes (possibility to zoom in/out the audio signal)
    + File explorer included in all modes, to facilitate the navigation between files
    + New Share mode to share recordings between devices (by Bluetooth, Mail, NFC if available)
    + French and German languages available. In addition to English, the application now supports French and German languages. Lig-Aikuma uses by default the language of the phone/tablet.
    + New icons, more consistent to discriminate all type of files (audio, text, image, video)
  • Conceptual upgrade:
    + New name for the root project: ligaikuma ?> /! Henceforth, all data will be stored into this directory instead of ?aikuma? (in the previous versions of the app). This change doesn?t have compatibility issues. In the file explorer of the mode, the default position is this root directory. Just go back once with the left grey arrow (on the lower left of the screen) and select the ?aikuma? directory to access to your old recordings
    + Generation of a PDF consent form (from informations filled in the metadata form) that can be signed by linguist and speaker thanks to a pdf annotation tool (like Adobe Fill & Sign mobile app)
    + Generation of a CSV file which can be imported in Elan software: it will automatically create segmented tier, as it was done during a respeaking or a translation session. It will also mention by a ?non-speech? label that a segment has no speech.
    + Géolocalisation of the recordings
    + Respeak an elicit file: it is now possible to use in Respeaking or Translation mode an audio file initially recorded in Elicitation mode
  • Structural upgrade:
    + Undo button on Elicitation to erase/redo the current recording
    + Improvement session backup on Elicitation
    + Non-speech button in Respeaking and Translation modes to indicate by a comment that the segment does not contain speech (but noise or silent for instance)
    + Automatic speaker profile creation to quickly fill in the metadata infos if several sessions with a same speaker
Best regards,

Elodie Gauthier & Laurent Besacier

5-3-7Python Library
Nous sommes heureux d'annoncer la mise à disposition du public de la
première bibliothèque en langage Python pour convertir des nombres écrits en
français en leur représentation en chiffres.
L'analyseur est robuste et est capable de segmenter et substituer les expressions
de nombre dans un flux de mots, comme une conversation par exemple. Il reconnaît les différentes
variantes de la langue (quantre-vingt-dix / nonante?) et traduit aussi bien les
ordinaux que les entiers, les nombres décimaux et les séquences formelles (n° de téléphone, CB?).
Nous espérons que cet outil sera utile à celles et ceux qui, comme nous, font du traitment
du langage naturel en français.
Cette bibliothèque est diffusée sous license MIT qui permet une utilisation très libre.
Romuald Texier-Marcadé

 Organisation  Events   Membership   Help 
 > Board  > Interspeech  > Join - renew  > Sitemap
 > Legal documents  > Workshops  > Membership directory  > Contact
 > Logos      > FAQ
       > Privacy policy

© Copyright 2024 - ISCA International Speech Communication Association - All right reserved.

Powered by ISCA