ALFFA partners participate to the Global Learning XPRIZE

ALFFA partners participate to the Global Learning XPRIZE:
  • The project is participating in a competition called Global Learning XPRIZE: http://learning.xprize.org
  • The solutions will be delivered to XPRIZE for judging November 2016. Then there will be a 6-month judging period, and the 5 best projects will receive 1.000.000 USD and enter into a final round of field testing.
  • The children will use the Android tablets autonomously for 18 months, and the goal is to enable them to learn how to read within this period.
  • Swahili TTS and ASR are developed for this specific use case
  • The Calculator made by Voxygen is used

SLTU 2016 Panel Discussion

During SLTU 2016 workshop in Indonesia, various institutions that have been working on under-resourced languages came and sat together to discuss the “Future of Under-Resourced Languages”.

The SLTU 2016_Panelist List as well as the slides presented there are gathered on this page…

Introduction: Joseph Mariani (LIMSI, France) and Sakriani Satki (NAIST, Japan) – Slides

SLTU Board: Laurent Besacier (LIG, France) – Slides
Zero Resource Speech Challenge: Emmanuel Dupoux (EHESS, France) – Slides
Babel Project: Stavros Tsakalidis (BBN, USA) – Slides
UNESCO: Ming Kuok Lim (UNESCO, Indonesia) – Slides
Academia: Steven Bird (Uni-Melbourne, Australia) – LetterRead
Industry: Martin Jansche (Google) – Slides
Open-Source Software: Kaldi: Sanjeev Khudanpur (JHU, USA)
Country Representatives: Mirna Adriani (UI, Indonesia) Slides – Shyam Agrawal (KIIT, India) Slides – Alexey Karpov (SPIIRAS, Russia) Slides – Charl van Heerden (NWU, South Africa) Slides – Tran Do Dat (MICA, Vietnam).

 

 

Laurent Besacier in the ELRA Workgroup on Less-Resourced Languages (LRL) Technical Committee.

Laurent Besacier is now in the ELRA Workgroup on Less-Resourced Languages (LRL) Technical Committee.

The ELRA Workgroup on Less-Resourced Languages (LRL) intends to bring together a number of professionals involved in the development of language resources and technologies for less-resourced languages. The Worksgroup’s main objective is to build a community that not only supports linguistic diversity through technology and ICT but also commits to increase less-resourced languages (regional, minority, or endangered) chances to survive the digital world.

ALFFA ASR Resources now mirrored on OpenSLR

ALFFA ASR Resources are now mirrored on OpenSLR.

http://www.openslr.org/25/

OpenSLR is a site devoted to hosting speech and language resources, such as training corpora for speech recognition, and software related to speech recognition. OpenSLR is an initiative from John Hopkins University (JHU).

LIG organizes TALAF 2016 Workshop

http://talaf.imag.fr/2016/

TALAf workshops take place every two years. The first workshop was held during the JEP-TALN-RÉCITAL 2012 conference on June 8, 2012 in Grenoble (see proceedings: http://aclweb.org/anthology//W/W12/#1300). The second one took place during the TALN 2014 conference on July 1, 2014 in Marseille (see proceedings: http://www.taln2014.org/site/actes-en-ligne/actes-en-ligne-ateliers/).

The third edition of TALAf will be held during the JEP-TALN-RECITAL conference, on July 4, 2016 at INALCO in Paris.

The roles of TALAf workshop are:

  • bring together researchers in the field through meetings at the workshop but also with the talaf@imag.fr mailing list;

  • pooling knowledge using open source tools, standards (ISO, Unicode), and publishing the resources produced with an open license (Creative Commons) to avoid including the loss of information when a project stop and can not be resumed immediately for lack of resources;

  • develop a set of best practices based on the experience of researchers ; set up simple efficient methodologies based on free or very cheap software for the development of resources, exchange about techniques that can avoid the use of non-existent resources and finally tavoid loss of time and energy.

TALAf workshops are supported by the non-profit organisation “Lexicologie Terminologie Traduction”: http://www.ltt.auf.org/index.php

PROGRAMME COMMITTEE

Martine Adda-Decker (CNRS-LPP & LIMSI, Paris, France)

Laurent Besacier (LIG, Grenoble, France)

Sokhna Bao Diop (Université Gaston Berger, St Louis du Sénégal, Sénégal)

Philippe Bretier (Voxygen, Pleumeur-Bodou, France)

Khalid Choukri (ELDA, Paris, France)

Mame Thierno Cissé (ARCIV, Université Cheikh Anta Diop, Dakar, Sénégal)

Chantal Enguehard (LINA, Nantes, France)

Núria Gala (LIF, Marseille, France)

Modi Issouf (Ministère de l’Éducation, Niamey, Niger)

Fary Silate Ka (IFAN, Université Cheikh Anta Diop, Dakar, Sénégal)

Mathieu Mangeot (LIG, Grenoble, France)

Chérif Mbodj, (Centre de Linguistique Appliquée de Dakar, Sénégal)

Kamal Naït-Zerrad (INALCO, Paris, France)

El Hadj Mamadou Nguer (Université Gaston Berger, St Louis du Sénégal, Sénégal)

Donald Osborn (Bisharat, ltd.)

Francois Pellegrino, (DDL, Lyon, France)

Olivier Rosec (Voxygen, Pleumeur-Bodou, France)

Fatiha Sadat (UQAM, Montréal, Canada)

Aliou Ngoné Seck (FLSH, Université Cheikh Anta Diop, Dakar, Sénégal)

Emmanuel Schang (Université d’Orléans, Orléans, France)

Gilles Sérasset (LIG, Grenoble, France)

Max Silberztein (ELLIADD, Université de Franche-Comté, Besançon, France)

Sylvie Voisin (DDL, Lyon, France)

Valentin Vydrin (LLACAN-INALCO, Paris, France)

IMPORTANT DATES

– Submission deadline: 24 April 2016

– Notification of acceptance: 11 May 2016

– Final Submission Deadline: 1 June 2016

– Workshop: 4 July 2016

Kaldi-based ASR system for Wolof now available to the public

All the data and scripts to build a complete ASR system for Wolof (based on Kaldi toolkit) is made available to the public on GitHub.

Wolof repository is composed of the kaldi recipes + ressources (the transcribed speech corpus with audio, pronunciation lexicon and LM).

More details on Wolof ASR (read speech transcription, WER=20.0% on cleaned dev data) can be found on https://github.com/besacier/ALFFA_PUBLIC/blob/master/ASR/WOLOF/README

SLTU 2016 paper is now online

This paper deals with ASR for two languages: Hausa and Wolof. Their common characteristic is to appear with vowel length contrast. In other words, two versions (short / long) of a same vowel exist in the phoneme inventory of the language. We expect that taking into account this contrast in ASR models might help and this is what we investigate in this pilot study. The experimental results show that while both approaches (vowel length contrast modeling or not) lead to similar results, their combination allows to slightly improve ASR performance. As a by-product of ASR system design, we also show that the acoustic models obtained allow a large scale analysis of vowel length contrast for phonetic studies.

See paper: ASR for African Languages With Vowel Length Contrast

LREC 2016 paper on ALFFA is now online

This article presents the data collected and ASR systems developped for 4 sub-saharan african languages (Swahili, Hausa, Amharic and Wolof). To illustrate our methodology, the focus is made on Wolof (a very under-resourced language) for which we designed the first ASR system ever built in this language. All data and scripts are available online on our github repository.

See paper: Collecting Resources in Sub-Saharan African Languages for Automatic Speech Recognition: a Case Study of Wolof

Interspeech 2016 Special Session

Sub-Saharan African languages : from speech fundamentals to applications

Organizers

Martine Adda-Decker1 (madda@limsi.fr) – CNRS – LPP and LIMSI, France.

Laurent Besacier2 (laurent.besacier@imag.fr) – Univ. Grenoble-Alpes, France – LIG laboratory.

Marelie Davel3 (marelie.davel@nwu.ac.za) – North-West University, Vanderbijlpark, South Africa.

Larry Hyman4 (hyman@berkeley.edu) – Department of Linguistics, University of California, Berkeley.

Martin Jansche5 (mjansche@google.com) – Google, London, UK.

Francois Pellegrino6 (francois.pellegrino@univ-lyon2.fr) – CNRS – DDL Lyon, France.

Olivier Rosec7 (olivier.rosec@voxygen.fr) – Voxygen SAS,- Pleumeur-Bodou, France.

Sebastian Stüker8 (sebastian.stueker@kit.edu) – Karlsruhe Institute of Technology (KIT), Germany.

Martha Tachbelie Yifiru9 (martha.yifiru@aau.edu.et) – School of Information Science, Addis Ababa University, Ethiopia.

Motivation

The number of languages spoken in Africa ranges from 1,500 to 2,500, depending on estimates and definitions. Africa is a place of rich — if not the richest — linguistic diversity which is often related to Africa’s being the cradle of humanity. Political explanations also go in favor of high linguistic diversity: during the 19th century, the many ethnic groups existing in Africa were divided into different countries according to colonial borders, resulting in cross-border languages. Multilingualism is thus ubiquitous in Sub-Saharan African societies.

In recent years, many institutions and organizations have been created to develop and standardize spoken and written resources in many languages (not only the higher-resourced ones), to provide multilingual access to digital resources and to promote national languages in the digital space. Coming to economical considerations, a major challenge of the digital market in Africa is to ensure that online applications and services become accessible to a wide range of people, be they literate or not. For the latter ones, applications should be able to talk and listen to Africans in the true sense of the words and several UNESCO reports make explicit reference to text to speech synthesis (TTS) and automatic speech recognition (ASR) as a technological facilitator. People’s access to ICT is done mainly through mobile phones and the need for voice services can be found in all sectors, from high priority (health, food, information) to more fun and leisure (sports, games, social media) areas. Beyond speech technologies, language documentation is another priority pointed out by UNESCO, since some languages in the Sub-Saharan Africa are seriously endangered with the risk of disappearing in the coming years.

However, progress in speech and language technologies in African languages suffers from a lack in linguistic knowledge and basic NLP resources. Now is the right time to bring speech technology and linguistic researchers together for the benefit of sharing fundamental linguistic knowledge and language resources, as well as technological, societal and economical progress.

To address these goals (language documentation, development of speech and language resources and technologies in African languages), inter-disciplinary research is required: from fundamentals of speech analysis (acoustic-phonetic, tonal and other linguistic descriptions, dialectology and typology) to speech technologies (ASR, TTS…). Speech and language technologies may also contribute to open up new perspectives in empirical linguistic research with new tools and approaches speeding up corpus and metadata production, language documentation of poorly studied languages, treating unwritten languages, thus contributing to the struggle against loss of language diversity in Africa.

To summarize, this special session aims at gathering researchers in speech technology (automatic speech recognition, text-to-speech synthesis…) and researchers in linguistics (field linguists, phoneticians…) working in language documentation and fundamentals of speech science. We believe that such a partnership is particularly important for Sub-Saharan African languages which tend to remain under-resourced, under-documented and often also un-written. While a minimum language description kit required to develop speech technologies cannot be obtained without linguistic research, automatic speech technology offers novel opportunities and tools to linguists for speeding up investigations in a wide range of topics in phonetics, phonology, tonology, morphology…

Possible contributors to the special session

Many researchers / institutions could submit a paper to this special session. In addition to the organizers of this session, the potential contributors can be found among the following projects, academic or industrial research groups :

  • Academic researchers working in Africa on their mother tongue language (such a session would be also a great opportunity to increase the participation of African researchers to Interspeech and to ISCA association),

  • Industrial research centers developing ASR and TTS for many languages (Google, Amazon, Microsoft, Facebook, etc.),

  • BULB project (Breaking the Unwritten Language Barrier) : KIT, LIMSI, LIG, ZAS, LPP, LLACAN,

  • BABEL project (IARPA program) : many partners from US, Europe and South-Africa,

  • ALFFA project (African Languages in the Field – Speech Fundamentals and Automation) : LIG, DDL, LIA, Voxygen SAS

Session format

The session will consist of very-short oral presentations (2mn booster) followed by a poster-session. The best papers will be encouraged to be submitted to a special issue in a journal (Speech Communication or Computer Speech and Language).

Call for papers

Prospective authors are invited to submit original papers in areas related to Sub-Saharan African languages science and technology including but not limited to :

  • Speech recognition for Sub-Saharan African languages and dialects

  • Text-to-speech synthesis for Sub-Saharan African languages and dialects

  • Cross-lingual and multi-lingual acoustic and lexical modeling

  • Applications of spoken language technologies for the African continent

  • Phonetic and linguistic studies in Sub-Saharan African languages

  • Zero resource speech technologies: unsupervised discovery of linguistic units

  • Language documentation for endangered languages of Africa

  • Machine-assisted annotation of speech and laboratory phonology

  • Resource production in African languages: text and speech corpora, lexica, language models

1Research Director at CNRS. Inter-disciplinary research in speech technologies and corpus phonetics.

2Founder of the bi-annual SLTU workshop series (Spoken Language Technologies for Under-resourced languages). Research on spoken language technologies for low-resourced (including Sub-Saharan African) languages.

3Director of the Multilingual Speech Technologies research group at North-West University in South Africa. Language technology development for under-resourced languages, with a focus on Southern African languages.

4Professor of Linguistics in the Department of Linguistics at the University of California at Berkeley. He is a specialist in phonology, typology, and African linguistics, particularly Bantu and other Niger-Congo.

5Software Engineer, Google Research for Low Resource Languages, Google London.

6Former director of the ‘Dynamique Du Langage’ Laboratory (DDL) and coordinator of the ‘laboratory of excellence’ ASLAN (Advanced Studies on Language complexity).

7R&D director of Voxygen SAS. Specialist of Text-to-speech synthesis. Current developments and interests involve Sub-Saharan African languages.

8Coordinator of the ANR-DFG BULB project (Breaking the Unwritten Language Barrier). Research on spoken language technologies at KIT.

9Head of the School of Information Sciences, Addis Ababa University (Ethiopia). ASR and NLP for Amharic lang.

Next ALFFA meeting on 26 jan 2016 in Grenoble

Next ALFFA meeting will be held on 26 jan 2016 in Grenoble…

Schedule (in French)

9h-9h30 Accueil/Café

9h30-10h Retour sur évaluation mi-parcours ANR & aspect contractuels éventuels

10h-12h Bilan activités par partenaire

LIG, VOX, DDL, LIA

12h-13h Repas (plateaux)

13h-14h Objectifs à court terme

-livre blanc Hausa à finaliser

-soumissions prevues à interspeech 2016 (session spéciale langues africaines)

-atelier Talaf pendant JEP-TALN 2016 début Juillet

14h-16h Objectifs année 3

-Langues visées pour livres blancs, ASR et TTS

-Collecte de données associées

-Discussion sur choix d’un nouveau “use case”

-Autres points éventuels

 

Top