Odyssey Speaker and Language Recognition Workshop 2012, the workshop of ISCA SIG Speaker and Language Characterization, was held in Singapore on 25-28 June 2012. Odyssey 2012 is glad to announce that its video recordings have been included in the ISCA Video Archive. http://www.isca-speech.org/iscaweb/index.php/archive/video-archive

Top

Techniques for Noise Robustness in Automatic Speech Recognition
Tuomas Virtanen, Rita Singh, Bhiksha Raj (editors)
ISBN: 978-1-1199-7088-0
Publisher: Wiley

Automatic speech recognition (ASR) systems are finding increasing use in everyday life. Many of the commonplace environments where the systems are used are noisy, for example users calling up a voice search system from a busy cafeteria or a street. This can result in degraded speech recordings and adversely affect the performance of speech recognition systems. As the use of ASR systems increases, knowledge of the state-of-the-art in techniques to deal with such problems becomes critical to system and application engineers and researchers who work with or on ASR technologies. This book presents a comprehensive survey of the state-of-the-art in techniques used to improve the robustness of speech recognition systems to these degrading external influences.

Key features:

*Reviews all the main noise robust ASR approaches, including signal separation, voice activity detection, robust feature extraction, model compensation and adaptation, missing data techniques and recognition of reverberant speech.
*Acts as a timely exposition of the topic in light of more widespread use in the future of ASR technology in challenging environments.
*Addresses robustness issues and signal degradation which are both key requirements for practitioners of ASR.
*Includes contributions from top ASR researchers from leading research units in the field

Top

Understanding Prosody: The Role of Context, Function and Communication

Ed. by Niebuhr, Oliver

Series:Language, Context and Cognition 13,   De Gruyter

http://www.degruyter.com/view/product/186201?format=G or http://linguistlist.org/pubs/books/get-book.cfm?BookID=63238

The volume represents a state-of-the-art snapshot of the research on prosody for phoneticians, linguists and speech technologists. It covers well-known models and languages. How are prosodies linked to speech sounds? What are the relations between prosody and grammar? What does speech perception tell us about prosody, particularly about the constituting elements of intonation and rhythm? The papers of the volume address questions like these with a special focus on how the notion of context-based coding, the knowledge of prosodic functions and the communicative embedding of prosodic elements can advance our understanding of prosody.

Top

 Albert Di Cristo: « La Prosodie de la Parole : Une Introduction », Editions de Boeck-Solal (296 p). 

Sommaire : 

Avant –propos, Introduction, ;

 Ch.1 : Eléments de définition ; 

 Ch 2. Situation de la prosodie dans le champ des sciences du langage et dans l’étude de la communication ; 

Ch 3. La prosodie sur les deux versants de la communication orale interindividuelle (production et compréhension) ; 

Ch 4. La prosodie et le cerveau ;

 Ch 5. La matérialité de la prosodie ; 

Ch 6. Les niveau d’analyse et de représentation de la prosodie ; 

Ch 7. Les théories, les modèles de la prosodie et leurs appareils formels ;

 Ch 8 La fonctionnalité plurielle de la prosodie ; 

Ch 9. Les relations de la prosodie avec les sens ; 

Epilogue. 

Suggestions de lecture ;

 Index des termes ; 

Index des noms propres.

Top

Björn Schuller, Anton Batliner Computational Paralinguistics: Emotion, Affect and Personality in Speech and Language Processing Wiley, ISBN: 978-1-119-97136-8, 344 pages, November 2013 Description - This book presents the methods, tools and techniques that are currently being used to recognise (automatically) the affect, emotion, personality and everything else beyond linguistics (‘paralinguistics’) expressed by or embedded in human speech and language. - It is the first book to provide such a systematic survey of paralinguistics in speech and language processing. The technology described has evolved mainly from automatic speech and speaker recognition and processing, but also takes into account recent developments within speech signal processing, machine intelligence and data mining. - Moreover, the book offers a hands-on approach by integrating actual data sets, software, and open-source utilities which will make the book invaluable as a teaching tool and similarly useful for those professionals already in the field. Key features: - Provides an integrated presentation of basic research (in phonetics/linguistics and humanities) with state-of-the-art engineering approaches for speech signal processing and machine intelligence. - Explains the history and state of the art of all of the sub-fields which contribute to the topic of computational paralinguistics. - Covers the signal processing and machine learning aspects of the actual computational modelling of emotion and personality and explains the detection process from corpus collection to feature extraction and from model testing to system integration. - Details aspects of real-world system integration including distribution, weakly supervised learning and confidence measures. - Outlines machine learning approaches including static, dynamic and context-sensitive algorithms for classification and regression. - Includes a tutorial on freely available toolkits, such as the open-source ‘openEAR’ toolkit for emotion and affect recognition co-developed by one of the authors, and a listing of standard databases and feature sets used in the field to allow for immediate experimentation enabling the reader to build an emotion detection model on an existing corpus. Links: - The book: http://eu.wiley.com/WileyCDA/WileyTitle/productCd-1119971365.html - Table of Contents (pdf): http://media.wiley.com/product_data/excerpt/65/11199713/1119971365-16.pdf - Chapter01 (pdf): http://media.wiley.com/product_data/excerpt/65/11199713/1119971365-14.pdf 

Top

 *****************************************************************
    ELRA - Language Resources Catalogue - Update  December  2013
    *****************************************************************

We are happy to announce that 1 new Speech Language Resource and 1 new Written Corpus are now available in our catalogue. 

ELRA-S0365 aGender
aGender contains speech sample recordings over public telephone lines with read and (semi-)spontaneous speech. Native German speakers called a voice portal from their private phone, and read text + answered some open questions. The corpus contains the voices of 945 German speakers (approx. minimum of 100 speakers per class), each delivering 18 speech items in up to six different sessions.
For more information, see: http://catalog.elra.info/product_info.php?products_id=1214

ELRA-W0074 Amharic-English bilingual corpus
The Amharic-English bilingual corpus contains parallel text from legal and news domains in Amharic script, in transliterated form and in English. The size of the corpus is of 232,653 words in Amharic and 291,701 in English.
For more information, see: http://catalog.elra.info/product_info.php?products_id=1215

For more information on the catalogue, please contact Valérie Mapelli mailto:mapelli@elda.org

Visit our On-line Catalogue: http://catalog.elra.info
Visit the Universal Catalogue: http://universal.elra.info
Archives of ELRA Language Resources Catalogue Updates: http://www.elra.info/LRs-Announcements.html

Top

ELRA releases free Language Resources
***************************************************

Anticipating users’ expectations, ELRA has decided to offer a large number of resources for free for Academic research use. Such an offer consists of several sets of speech, text and multimodal resources that are regularly released, for free, as soon as legal aspects are cleared. A first set was released in May 2012 at the occasion of LREC 2012. A second set is now being released.

Whenever this is permitted by our licences, please feel free to use these resources for deriving new resources and depositing them with the ELRA catalogue for community re-use.

Over the last decade, ELRA has compiled a large list of resources into its Catalogue of LRs. ELRA has negotiated distribution rights with the LR owners and made such resources available under fair conditions and within a clear legal framework. Following this initiative, ELRA has also worked on LR discovery and identification with a dedicated team which investigated and listed existing and valuable resources in its 'Universal Catalogue', a list of resources that could be negotiated on a case-by-case basis. At LREC 2010, ELRA introduced the LRE Map, an inventory of LRs, whether developed or used, that were described in LREC papers. This huge inventory listed by the authors themselves constitutes the first 'community-built' catalogue of existing or emerging resources, constantly enriched and updated at major conferences.

Considering the latest trends on easing the sharing of LRs, from both legal and commercial points of view, ELRA is taking a major role in META-SHARE, a large European open infrastructure for sharing LRs. This infrastructure will allow LR owners, providers and distributors to distribute their LRs through an additional and cost-effective channel.

To obtain the available sets of LRs, please visit the web page below and follow the instructions given online:
http://www.elra.info/Free-LRs,26.html

Top

In this newsletter:

Spring 2014 LDC Data Scholarship recipients!

Membership fee savings and publications pipeline

New LDC website enhancements coming soon

New publications:

GALE Arabic-English Parallel Aligned Treebank -- Broadcast News Part 2

King Saud University Arabic Speech Database

NIST 2012 Open Machine Translation (OpenMT) Progress Test Five Language Source

Spring 2014 LDC Data Scholarship recipients!

LDC is pleased to announce the student recipients of the Spring 2014 LDC Data Scholarship program!  This program provides university students with access to LDC data at no-cost. Students were asked to complete an application which consisted of a proposal describing their intended use of the data, as well as a letter of support from their thesis adviser. We received many solid applications and have chosen two proposals to support.   The following students will receive no-cost copies of LDC data:

• Skye Anderson ~ Tulane University (USA), BA candidate, Linguistics.  Skye has been awarded a copy of LDC Standard Arabic Morphological Analyzer (SAMA) Version 3.1 for her work in author profiling.
• Hao Liu ~ University College London (UK), PhD candidate, Speech, Hearing and Phonetic Sciences.  Hao has been awarded a copy of Switchboard-1 Release 2, and NXT Switchboard Annotations for his work in prosody modeling.

Membership fee savings and publications pipeline

Members can still save on 2014 membership fees, but time is running out. Any organization which joins or renews membership for 2014 through Monday, March 3, 2014, is entitled to a 5% discount. Organizations which held membership for MY2013 can receive a 10% discount on fees provided they renew prior to March 3, 2014.

Planned publications for this year include:

• 2009 NIST Language Recognition Evaluation ~  development data from VOA broadcast and CTS telephone speech in target and non-target languages.
• ETS Corpus of Non-Native Written English ~ contains 1100 essays written for a college-entrance test sampled from eight prompts (i.e., topics)  with score levels (low/medium/high) for each essay.
• GALE data ~ including Word Alignment, Broadcast Speech & Transcripts, Parallel Text, Parallel Aligned Treebanks in Arabic, Chinese, and English.
• Hispanic Accented English ~ contains approximately 30 hours of spontaneous speech and read utterances from non-native speakers of English with corresponding transcripts.
• Multi-Channel Wall Street Journal Audio-Visual Corpus (MC-WSJ-AV) ~  re-recording of parts of the WSJCAM0 using a number of microphones as well as three recording conditions resulting in 18-20 channels of audio per recording.
• TAC KBP Reference Knowledge Base ~  TAC KBP aims to develop and evaluate technologies for building and populating knowledge bases (KBs) about named entities from unstructured text.  KBP systems must either populate an existing reference KB, or else build a KB from scratch. The reference KB for is based on a snapshot of English Wikipedia snapshot from October 2008 and contains a set of entities, each with a canonical name and title for the Wikipedia page, an entity type, an automatically parsed version of the data from the infobox in the entity's Wikipedia article, and a stripped version of the text of the Wiki article.
• USC-SFI MALACH Interviews and Transcripts Czech ~ developed by The University of Southern California's Shoah Foundation Institute (USC-SFI) and the University of West Bohemia as part of the MALACH (Multilingual Access to Large Spoken ArCHives) Project. It contains approximately 143 hours of interviews from 420 interviewees along with transcripts and other documentation.

New LDC website enhancements coming soon

Look for LDC’s new website enhancements in the coming weeks. We've revamped our membership services to make it easier than ever for you to manage your membership and access data more quickly.

New publications

(1) GALE Arabic-English Parallel Aligned Treebank -- Broadcast News Part 2 was developed by LDC and contains 141,058 tokens of word aligned Arabic and English parallel text with treebank annotations. This material was used as training data in the DARPA GALE (Global Autonomous Language Exploitation) program.

Parallel aligned treebanks are treebanks annotated with morphological and syntactic structures aligned at the sentence level and the sub-sentence level. Such data sets are useful for natural language processing and related fields, including automatic word alignment system training and evaluation, transfer-rule extraction, word sense disambiguation, translation lexicon extraction and cultural heritage and cross-linguistic studies. With respect to machine translation system development, parallel aligned treebanks may improve system performance with enhanced syntactic parsers, better rules and knowledge about language pairs and reduced word error rate.

In this release, the source Arabic data was translated into English. Arabic and English treebank annotations were performed independently. The parallel texts were then word aligned. The material in this corpus corresponds to a portion of the Arabic treebanked data in Arabic Treebank - Broadcast News v1.0 (LDC2012T07).

The source data consists of Arabic broadcast news programming collected by LDC in 2007 and 2008. All data is encoded as UTF-8. A count of files, words, tokens and segments is below.

The purpose of the GALE word alignment task was to find correspondences between words, phrases or groups of words in a set of parallel texts. Arabic-English word alignment annotation consisted of the following tasks:

• Identifying different types of links: translated (correct or incorrect) and not translated (correct or incorrect)
• Identifying sentence segments not suitable for annotation, e.g., blank segments, incorrectly-segmented segments, segments with foreign languages
• Tagging unmatched words attached to other words or phrases

GALE Arabic-English Parallel Aligned Treebank -- Broadcast News Part 2 is distributed via web download.

2014 Subscription Members will automatically receive two copies of this data on disc. 2014 Standard Members may request a copy as part of their 16 free membership corpora.  Non-members may license this data for US$1750.

(2) King Saud University Arabic Speech Database was developed by King Saud University and contains 590 hours of recorded Arabic speech from male and female speakers. The utterances include read and spontaneous speech. The recordings were conducted in varied environments representing quiet and noisy settings.

The corpus was designed principally for speaker recognition research. The speech sources are sentences, word lists, prose and question and answer sessions. Read speech text includes the following:

• Sets of sentences devised to cover allophones of each phoneme, phonetic balance, and differentiation of accents.
• Word lists developed to minimize missing phonemes and to represent nasals fricatives, commonly used words, and numbers.
• Two paragraphs, one from the Quran and another from a book, selected because they included all letters of the alphabet and were easy to read.

Spontaneous speech was captured through question and answer sessions between participants and project team members. Speakers responded to questions on general topics such as the weather and food.

Each speaker was recorded in three different environments: a sound proof room, an office, and a cafeteria. The recordings were collected via microphone and mobile phone and averaged between 16-19 minutes. The data was verified for missing recordings, problems with the recording system or errors in the recording process.

King Saud University Arabic Speech Database is distributed on one hard disk.

2014 Subscription Members will receive a copy of this data provided that they have completed the User License Agreement. 2014 Standard Members may request a copy as part of their 16 free membership corpora.  Non-members may license this data for US$2000.

(3) NIST 2012 Open Machine Translation (OpenMT) Progress Test Five Language Source was developed by NIST Multimodal Information Group. This release contains the evaluation sets (source data and human reference translations), DTD, scoring software, and evaluation plan for the OpenMT 2012 test for Arabic, Chinese, Dari, Farsi, and Korean to English on a parallel data set. The set is based on a subset of the Arabic-to-English and Chinese-to-English progress tests from the OpenMT 2008, 2009 and 2012 evaluations with new source data created by humans based on the English reference translation. The package was compiled, and scoring software was developed, at NIST, making use of newswire and web data and reference translations developed by the Linguistic Data Consortium  and the Defense Language Institute Foreign Language Center.

The objective of the OpenMT evaluation series is to support research in, and help advance the state of the art of, machine translation (MT) technologies -- technologies that translate text between human languages. Input may include all forms of text. The goal is for the output to be an adequate and fluent translation of the original. The 2012 task included the evaluation of five language pairs: Arabic-to-English, Chinese-to-English, Dari-to-English, Farsi-to-English and Korean-to-English in two source data styles. For general information about the NIST OpenMT evaluations, refer to the NIST OpenMT website.

This evaluation kit includes a single Perl script (mteval-v13a.pl) that may be used to produce a translation quality score for one (or more) MT systems. The script works by comparing the system output translation with a set of (expert) reference translations of the same source text. Comparison is based on finding sequences of words in the reference translations that match word sequences in the system output translation.

This release consists of 20 files, four for each of the five languages, presented in XML with an included DTD. The four files are source and reference data in the following two styles:

• English-true: an English-oriented translation this requires that the text read well and not use any idiomatic expressions in the foreign language to convey meaning, unless absolutely necessary.
• Foreign-true: a translation as close as possible to the foreign language, as if the text had originated in that language.

NIST 2012 Open Machine Translation (OpenMT) Progress Test Five Language Source is distributed via web download.

2014 Subscription Members will automatically receive two copies of this data on disc. 2014 Standard Members may request a copy as part of their 16 free membership corpora.  Non-members may license this data for US$150.

Top

Appen ButlerHill 

A global leader in linguistic technology solutions

RECENT CATALOG ADDITIONS—MARCH 2012

1. Speech Databases

1.1 Telephony

2. Pronunciation Lexica

Appen Butler Hill has considerable experience in providing a variety of lexicon types. These include:

 Pronunciation Lexica providing phonemic representation, syllabification, and stress (primary and secondary as appropriate)

 Part-of-speech tagged Lexica providing grammatical and semantic labels

 Other reference text based materials including spelling/mis-spelling lists, spell-check dictionar-ies, mappings of colloquial language to standard forms, orthographic normalization lists.

Over a period of 15 years, Appen Butler Hill has generated a significant volume of licensable material for a wide range of languages. For holdings information in a given language or to discuss any customized development efforts, please contact: sales@appenbutlerhill.com

4. Other Language Resources

 Morphological Analyzers – Farsi/Persian & Urdu

 Arabic Thesaurus

 Language Analysis Documentation – multiple languages

For additional information on these resources, please contact: sales@appenbutlerhill.com

5. Customized Requests and Package Configurations

Appen Butler Hill is committed to providing a low risk, high quality, reliable solution and has worked in 130+ languages to-date supporting both large global corporations and Government organizations.

We would be glad to discuss to any customized requests or package configurations and prepare a cus-tomized proposal to meet your needs.

                                                         www.appenbutlerhill.com

Top

We are happy to announce the release of DEMAND, a set of real-world
16-channel noise recordings designed for the evaluation of microphone
array processing techniques.

http://www.irisa.fr/metiss/DEMAND/

1.5 h of noise data were recorded in 18 different indoor and outdoor
environments and are available under the terms of the Creative Commons Attribution-ShareAlike License.

Joachim Thiemann (CNRS - IRISA)
Nobutaka Ito (University of Tokyo)
Emmanuel Vincent (Inria Nancy - Grand Est)

Top

Aide à la finalisation de corpus oraux ou multimodaux pour diffusion, valorisation et dépôt pérenne

Le consortium IRCOM de la TGIR Corpus et l’EquipEx ORTOLANG s’associent pour proposer une aide technique et financière à la finalisation de corpus de données orales ou multimodales à des fins de diffusion et pérennisation par l’intermédiaire de l’EquipEx ORTOLANG. Cet appel ne concerne pas la création de nouveaux corpus mais la finalisation de corpus existants et non-disponibles de manière électronique. Par finalisation, nous entendons le dépôt auprès d’un entrepôt numérique public, et l’entrée dans un circuit d’archivage pérenne. De cette façon, les données de parole qui ont été enrichies par vos recherches vont pouvoir être réutilisées, citées et enrichies à leur tour de manière cumulative pour permettre le développement de nouvelles connaissances, selon les conditions d’utilisation que vous choisirez (sélection de licences d’utilisation correspondant à chacun des corpus déposés).

Cet appel d’offre est soumis à plusieurs conditions (voir ci-dessous) et l’aide financière par projet est limitée à 3000 euros. Les demandes seront traitées dans l’ordre où elles seront reçues par l’ IRCOM. Les demandes émanant d’EA ou de petites équipes ne disposant pas de support technique « corpus » seront traitées prioritairement. Les demandes sont à déposer du 1^er septembre 2013 au 31 octobre 2013. La décision de financement relèvera du comité de pilotage d’IRCOM. Les demandes non traitées en 2013 sont susceptibles de l’être en 2014. Si vous avez des doutes quant à l’éligibilité de votre projet, n’hésitez pas à nous contacter pour que nous puissions étudier votre demande et adapter nos offres futures.

Pour palier la grande disparité dans les niveaux de compétences informatiques des personnes et groupes de travail produisant des corpus, L’ IRCOM propose une aide personnalisée à la finalisation de corpus. Celle-ci sera réalisée par un ingénieur IRCOM en fonction des demandes formulées et adaptées aux types de besoin, qu’ils soient techniques ou financiers.

Les conditions nécessaires pour proposer un corpus à finaliser et obtenir une aide d’IRCOM sont :

• Pouvoir prendre toutes décisions concernant l’utilisation et la diffusion du corpus (propriété intellectuelle en particulier).

• Disposer de toutes les informations concernant les sources des corpus et le consentement des personnes enregistrées ou filmées.

• Accorder un droit d’utilisation libre des données ou au minimum un accès libre pour la recherche scientifique.

Les demandes peuvent concerner tout type de traitement : traitements de corpus quasi-finalisés (conversion, anonymisation), alignement de corpus déjà transcrits, conversion depuis des formats « traitement de textes », digitalisation de support ancien. Pour toute demande exigeant une intervention manuelle importante, les demandeurs devront s’investir en moyens humains ou financiers à la hauteur des moyens fournis par IRCOM et ORTOLANG.

IRCOM est conscient du caractère exceptionnel et exploratoire de cette démarche. Il convient également de rappeler que ce financement est réservé aux corpus déjà largement constitués et ne peuvent intervenir sur des créations ex-nihilo. Pour ces raisons de limitation de moyens, les propositions de corpus les plus avancés dans leur réalisation pourront être traitées en priorité, en accord avec le CP d’IRCOM. Il n’y a toutefois pas de limite « théorique » aux demandes pouvant être faites, IRCOM ayant la possibilité de rediriger les demandes qui ne relèvent pas de ses compétences vers d’autres interlocuteurs.

Les propositions de réponse à cet appel d’offre sont à envoyer à ircom.appel.corpus@gmail.com. Les propositions doivent utiliser le formulaire de deux pages figurant ci-dessous. Dans tous les cas, une réponse personnalisée sera renvoyée par IRCOM.

Ces propositions doivent présenter les corpus proposés, les données sur les droits d’utilisation et de propriétés et sur la nature des formats ou support utilisés.

Cet appel est organisé sous la responsabilité d’IRCOM avec la participation financière conjointe de IRCOM et l’EquipEx ORTOLANG.

Pour toute information complémentaire, nous rappelons que le site web de l'Ircom (http://ircom.corpus-ir.fr) est ouvert et propose des ressources à la communauté : glossaire, inventaire des unités et des corpus, ressources logicielles (tutoriaux, comparatifs, outils de conversion), activités des groupes de travail, actualités des formations, ...

L'IRCOM invite les unités à inventorier leur corpus oraux et multimodaux - 70 projets déjà recensés - pour avoir une meilleure visibilité des ressources déjà disponibles même si elles ne sont pas toutes finalisées.

Le comité de pilotage IRCOM

Utiliser ce formulaire pour répondre à l’appel : Merci.

Réponse à l’appel à la finalisation de corpus oral ou multimodal

Nom du corpus :

Nom de la personne à contacter :

Adresse email :

Numéro de téléphone :

Nature des données de corpus :

Existe-t’il des enregistrements :

Quel média ? Audio, vidéo, autre…

Quelle est la longueur totale des enregistrements ? Nombre de cassettes, nombre d’heures, etc.

Quel type de support ?

Quel format (si connu) ?

Existe-t’il des transcriptions :

Quel format ? (papier, traitement de texte, logiciel de transcription)

Quelle quantité (en heures, nombre de mots, ou nombre de transcriptions) ?

Disposez vous de métadonnées (présentation des droits d’auteurs et d’usage) ?

Disposez-vous d’une description précise des personnes enregistrées ?

Disposez-vous d’une attestation de consentement éclairé pour les personnes ayant été enregistrées ? En quelle année (environ) les enregistrements ont eu lieu ?

Quelle est la langue des enregistrements ?

Le corpus comprend-il des enregistrements d’enfants ou de personnes ayant un trouble du langage ou une pathologie ?

Si oui, de quelle population s’agit-il ?

Dans un souci d’efficacité et pour vous conseiller dans les meilleurs délais, il nous faut disposer d’exemples des transcriptions ou des enregistrements en votre possession. Nous vous contacterons à ce sujet, mais vous pouvez d’ores et déjà nous adresser par courrier électronique un exemple des données dont vous disposez (transcriptions, métadonnées, adresse de page web contenant les enregistrements).

Nous vous remercions par avance de l’intérêt que vous porterez à notre proposition. Pour toutes informations complémentaires veuillez contacter Martine Toda martine.toda@ling.cnrs.fr ou à ircom.appel.corpus@gmail.com.

Top

Rhapsodie: un Treebank prosodique et syntaxique de français parlé

Nous avons le plaisir d'annoncer que la ressource Rhapsodie, Corpus de français parlé annoté pour la prosodie et la syntaxe, est désormais disponible sur http://www.projet-rhapsodie.fr/

Le treebank Rhapsodie est composé de 57 échantillons sonores (5 minutes en moyenne, au total 3h de parole, 33000 mots) dotés d’une transcription orthographique et phonétique alignées au son.

Il s'agit d’une ressource de français parlé multi genres (parole privée et publique ; monologues et dialogues ; entretiens en face à face vs radiodiffusion, parole plus ou moins interactive et plus ou moins planifiée, séquences descriptives, argumentatives, oratoires et procédurales) articulée autour de sources externes (enregistrements extraits de projets antérieurs, en accord avec les concepteurs initiaux) et internes. Nous tenons en particulier à remercier les responsables des projets CFPP2000, PFC, ESLO, C-Prom ainsi que Mathieu Avanzi, Anne Lacheret, Piet Mertens et Nicolas Obin.

Les échantillons sonores (wave & MP3, pitch nettoyé et lissé), les transcriptions orthographiques (txt), les annotations macrosyntaxiques (txt), les annotations prosodiques (xml, textgrid) ainsi que les metadonnées (xml & html) sont téléchargeables librement selon les termes de la licence Creative Commons Attribution - Pas d’utilisation commerciale - Partage dans les mêmes conditions 3.0 France.

Les annotations microsyntaxiques seront disponibles prochainement

 Les métadonnées sont également explorables en ligne grâce à un browser.

 Les tutoriels pour la transcription, les annotations et les requêtes sont disponibles sur le site Rhapsodie.

 Enfin, L’annotation prosodique est interrogeable en ligne grâce au langage de requêtes Rhapsodie QL.

 L'équipe Ressource Rhapsodie (Modyco, Université Paris Ouest Nanterre)

Sylvain Kahane, Anne Lacheret, Paola Pietrandrea, Atanas Tchobanov, Arthur Truong.

 Partenaires : IRCAM (Paris), LATTICE (Paris), LPL (Aix-en-Provence), CLLE-ERSS (Toulouse).

********************************************************

Rhapsodie: a Prosodic and Syntactic Treebank for Spoken French

We are pleased to announce that Rhapsodie, a syntactic and prosodic treebank of spoken French created with the aim of modeling the interface between prosody, syntax and discourse in spoken French is now available at   http://www.projet-rhapsodie.fr/

The Rhapsodie treebank is made up of 57 short samples of spoken French (5 minutes long on average, amounting to 3 hours of speech and a 33 000 word corpus) endowed with an orthographical phoneme-aligned transcription . 

The corpus is representative of different genres (private and public speech; monologues and dialogues; face-to-face interviews and broadcasts; more or less interactive discourse; descriptive, argumentative and procedural samples, variations in planning type).

The corpus samples have been mainly drawn from existing corpora of spoken French and partially created within the frame of theRhapsodie project. We would especially like to thank the coordinators of the  CFPP2000, PFC, ESLO, C-Prom projects as well as Piet Mertens, Mathieu Avanzi, Anne Lacheret and Nicolas Obin.

The sound samples (waves, MP3, cleaned and stylized pitch), the orthographic transcriptions (txt), the macrosyntactic annotations (txt), the prosodic annotations  (xml, textgrid) as well as the metadata (xml and html) can be freely downloaded under the terms of the Creative Commons licence Attribution - Noncommercial - Share Alike 3.0 France.

Microsyntactic annotation will be available soon.

The metadata are  searchable on line through a browser.

The prosodic annotation can be explored on line through the Rhapsodie Query Language.

The tutorials of transcription, annotations and Rhapsodie Query Language  are available on the site.

The Rhapsodie team (Modyco, Université Paris Ouest Nanterre :

Sylvain Kahane, Anne Lacheret, Paola Pietrandrea, Atanas Tchobanov, Arthur Truong.

Partners: IRCAM (Paris), LATTICE (Paris), LPL (Aix-en-Provence),CLLE-ERSS (Toulouse).

Top

We have created and made publicly available a dense audio-visual person-oriented ground-truth annotation of a feature movie (100 minutes long): “Hannah and her sisters” by Woody Allen.

The annotation includes

•          Face tracks in video (densely annotated, i.e., in each frame, and person-labeled)

•             Speech segments in audio (person-labeled)

•             Shot boundaries in video



The annotation can be useful for evaluating



•   Person-oriented video-based tasks (e.g., face tracking, automatic character naming, etc.)

•             Person-oriented audio-based tasks (e.g., speaker diarization or recognition)

•             Person-oriented multimodal-based tasks (e.g., audio-visual character naming)



Detail on Hannah dataset and access to it can be obtained there:

https://research.technicolor.com/rennes/hannah-home/

https://research.technicolor.com/rennes/hannah-download/



Acknowledgments:

This work is supported by AXES EU project: http://www.axes-project.eu/










Alexey Ozerov Alexey.Ozerov@technicolor.com

Jean-Ronan Vigouroux,

Louis Chevallier

Patrick Pérez

Technicolor Research & Innovation



 

Top

Text to         Speech Synthesis:
      over an hour of speech       synthesis samples from         1968 to 2001 by       25 French, Canadian, US , Belgian,       Swedish, Swiss systems
     
      33 ans de synthèse de la parole à         partir du texte: une promenade sonore (1968-2001)
        (33 years of Text to Speech Synthesis       in French : an audio tour (1968-2001)       )
      Christophe d'Alessandro
      Article published in         Volume 42 - No. 1/2001 issue of  Traitement       Automatique des Langues  (TAL,       Editions Hermes),         pp. 297-321.
     
      posted to:
      http://groupeaa.limsi.fr/corpus:synthese:start

Top

Press Release - Immediate - Paris, France, December 13, 2013

Establishing the International Standard Language Resource Number (ISLRN)

12 major NLP organisations announce the establishment of the ISLRN, a Persistent Unique Identifier, to be assigned to each Language Resource.

On November 18, 2013, 12 NLP organisations have agreed to announce the establishment of the International Standard Language Resource Number (ISLRN), a Persistent Unique Identifier, to be assigned to each Language Resource. Experiment replicability, an essential feature of scientific work, would be enhanced by such unique identifier. Set up by ELRA, LDC and AFNLP/Oriental-COCOSDA, the ISLRN Portal will provide unique identifiers using a standardised nomenclature, as a service free of charge for all Language Resource providers. It will be supervised by a steering committee composed of representatives of participating organisations and enlarged whenever necessary.

More information on ELRA and the ISLRN, please contact: Khalid Choukri choukri@elda.org

More information on ELDA, please contact: Hélène Mazo mazo@elda.org

ELRA

55-57, rue Brillat Savarin

75013 Paris (France)

Tel.: +33 1 43 13 33 33

Fax: +33 1 43 13 33 30

Top

Speechocean March 2014 update:

Speechocean: A global language resources and data services supplier

has over 500 large-scale databases available in 110+ languages and accents with the platform of desktop, in-car, telephony and tablet PC. Our data repository is enormous and diversified, which includes ASR Databases, TTS Databases, Lexica, Text Corpora, etc.

Speechocean is glad to announce that more resources have been released:

ASR Databases

Speechocean provides 110+ regional languages corpora, available in a variety of formats, situational styles, scene environments and platform systems, covering In-car speech recognition corpora, mobile phone speech recognition corpora, fixed-line speech recognition corpora, desktop speech recognition corpora, etc. This month we released more Asian languages databases which were made for the tuning and testing purpose of speech recognition systems for speech ASR applications.

1. In-Car

Chinese Mandarin Speech Recognition Database ---- (In-Car)-100 Speakers

ID: King-ASR-122

This database was collected in China Mainland. It contains the voices of 100 different native speakers (50 males, 50 females) who were balanced according to age(mainly 18 – 30（62），31 – 45（28）,46 – 60（10）), gender (Male 50%, Female 50%) and regional accents (Northern 60%, Wu 10%, Xiang 5%, Gan 5%, Kejia 5%, Min 5%, Cantonese 10%).

The script was specially designed to provide material for both training and testing of many classes of speech recognizers which contain 320 utterances covering 15 categories and 35 sub-categories for each speaker (for the detail script structure design, please see the technical document).
Each speaker was recorded under two environments in three variations (Parked, City Driving and Highway driving) with various kinds of recording conditions such as motor running, fan on/off, window up/down, etc. 320 utterances were recorded for each speaker under two environments and there are 200796 utterances recorded in total.

Each utterance is stored in a separate file and each signal file is accompanied by an ASCII SAM label file which contains the relevant descriptive information. A pronunciation lexicon with a phonemic transcription in SAMPA is also included. All the data was transcribed and labeled.

Japanese Speech Recognition Database ---- (In-Car)-800 Speakers

ID: King-ASR-125

This Japan In-car Speech database was collected in Japan and contains the voices of 800 different native speakers who were demographically balanced according to Age (16-30, 31-45, and 46-60), Gender (400±5% males, 400±5% females) and Dialectical Region. The script was specially designed to provide material for both training and testing of many classes of speech recognizers which contains 16 general categories and more than 50 specific sub-categories. Each speaker was recorded under three driving environments (parked, city driving and highway driving) with recording conditions such as fan on/off and window up/down. A total of 300 utterances were recorded for each speaker in two of three driving environments (150 utterances and 10 spontaneous utterances per environment).

Four high quality audio channels (C1: SHURE SM10A, C2: SENNHEISER ME104, F1: AKG Q400, F2: AKG Q400) and three popular cars in the country were used in this recording.

The speech data is stored as sequences of 16 kHz, 16 bit which is uncompressed and each prompted utterance is stored in a separate file and each signal file is accompanied by an ASCII SAM label file which contains the relevant descriptive information. A pronunciation lexicon with a phonemic transcription in SAMPA is also included. All the data was transcribed and labeled.

2. Telephony

Japanese Speech Recognition Database ---- Conversation (Telephony)-201 Speakers

ID: King-ASR-055

This Japanese Speech Recognition database was collected in Japan and contains the voices of 201 different native speakers who were demographic balanced according to age distribution (16-28,29-60), Gender, Dialectical Regions. The corpus contains 100 pairs of spontaneous dialog speech data which were from 201 speakers. Each pair of speech consists of 3 audio files: two of them from single speaker separately and the other is from the mixed channel. The three files were recorded simultaneously. The pure recording time of mixed channel is about 104.8 hours. 33 topics were contained in this database.

There are 7,009 audio files which were saved as uncompressed PCM files. All the speech data was transcribed and labeled.

1.3 Mobile

Korean Speech Recognition Database—(Mobile)--1023 Speakers

ID: King-ASR-137

The Korean mobile speech Recognition database which was collected in Korea, contains the voices of 1023 different native speakers (510±5%males, 513±5% females) who were balanced according to age (mainly 16 – 30,31 – 45,46 – 60), Gender and regional accents (for the details, please see the technical document).

The script was specially designed to provide material for both training and testing of many classes of speech recognizers which contain 15 categories and 35 sub-categories for each speaker (for the detail script structure design, please see the technical document).

Each speaker has recorded 300 utterances under two environments, one in a quiet session (Office/Home) and one in a noisy session (Garden/roadside/restaurant/bus). Each speaker has recorded 150 utterances and spontaneous sentences per session and totally 300 utterances were recorded by each speaker.

Popular mobiles in this country were used for collecting this data such as Samsung, Nokia, HTC, etc. The speech data is stored as sequences of 16 kHz, 16 bit and uncompressed.
Each utterance is stored in a separate file and each signal file is accompanied by an ASCII SAM label file which contains the relevant descriptive information.

A pronunciation lexicon with a phonemic transcription is also included. All the data was transcribed and labeled

Chinese Mandarin Speech Recognition Database---Sentences (Mobile) - (5048 Speakers)

ID: King-ASR-216

This database is a desktop speech database collected by Speechocean which is performed in a quiet environment in China. This database is one of our databases of Speech Data ----Mobile Project (SDM) which contains the database collections in 30 languages presently.

This database contains 1,514,028 sentences of Chinese Mandarin speech data which were from 5048 speakers which were recorded in a quiet environment. The pure recording length is about 2,268 hours. All speakers are native speakers from 14 typical dialectical cities covering seven main dialectical regions of China who were demographic balanced according to age distribution (16~30, 31~45, 46~60), Gender (2,584 Males and 2464 Females) and regional accents.

The script was specially designed to provide material for both training and testing of many classes of speech recognizers. The script of each speaker contains 300 sentences which were randomly selected from a pool of sentences specially designed. Each speaker will be recorded as naturally as possible in quiet environment through Popular Mobile Phones such as of iPhones, HTC Samsung, MOTO and etc. which cover the platforms of ios, android and window mobile.

The speech data are stored as sequences of 16 kHz, 16 bit and uncompressed PCM format. All the speech was manually transcribed and labeled. A pronunciation lexicon with a phonemic transcription in Pinyin is also included.

4. Desktop

Indonesian Speech Recognition Database ---- Sentences (Desktop)-200 Speakers

ID: King-ASR-061

This Indonesian Speech Recognition database was collected in Indonesia and contains the voices of 200 different native speakers who were demographic balanced according to age distribution (16–30, 31–45, 46–60) and Gender. It contains 239267 audio files with about 460.94 hours of recording.

Each speaker uttered 300 sentences in a quiet office room. The whole data has been proofread manually with precise data labeling.

Urdu Speech Recognition Database ----Sentences (Desktop)-200 Speakers

ID: King-ASR-063

This Urdu Speech Recognition database, which was collected in Pakistan, contains the voices of 200 native speakers who were demographic balanced according to age distribution (16–60), gender, dialectical Regions, there were 241,354 audio files which were saved as uncompressed PCM files. All the speech data was transcribed and labeled.

Vietnamese Speech Recognition Database ----Sentences (Desktop)-200 Speakers

ID: King-ASR-074

This Vietnamese Speech Recognition database, which was collected in Vietnam, contains the voices of 200 native speakers who were demographic balanced according to age distribution (16–60), Gender, Dialectical Regions, there were 263,204 audio files which were saved as uncompressed PCM files. All the speech data was transcribed and labeled.

2. TTS Databases

Speechocean licenses a variety of databases in more than 40 languages for speech synthesis broadcasting speech, emotional speech, etc. which can be used in different algorithms.

European Portuguese Speech Corpus for TTS (Female)

ID: King-TTS-017

The European Portuguese (pt-PT) Speech Corpus consists native Portuguese female professional broadcaster (Female, 32 years old) recorded in a studio with high SNR (>35dB) over two channels (Shure SM15 microphone and Electroglottography (EGG) sensor).

The Corpus includes the following sub-corpora:

1. Sentence sub-corpus: including 3000 short sentences (7~12 words) and 2000 sentences with normal length (13~20 words). Considering all kinds of linguistic phenomena, all sentences are extracted from the daily articles in Portugal, such as national and international news, papers in life, travel, and so on. The sentences with political/religious/obscene/pornographic words which might lead to negative emotions are carefully excluded.

2. Emotional sub-corpus: including 100 exclamatory sentences and 100 interrogative sentences which can be used for emotional TTS study;

3. Digit sub-corpus: including many kinds of digits data, such as isolated digit, connected digits with blocks, natural and ordinal number readings;

4. Expression sub-corpus: consists of general expressions, such as date, time, money and measure expression;

5. Spell sub-corpus: including characters in alphabet, Greek characters and general abbreviations;

All reading prompts are manually revised and prosody annotations were made according to real speech. All speech data are segmented and labeled on phone level. Pronunciation lexicon and pitch extract from EEG can also be provided based on demands

3. Text Corpora

Speechocean licenses many kinds of text corpora in many languages which is superb for language model training.

4. Lexica

Speechocean builds pronunciation lexica in many languages which can be licensed to customers.

Contact Information

Xianfeng Cheng

Business Manager of Commercial Department

Tel: +86-10-62660928; +86-10-62660053 ext.8080

Cell phone: +86 13681432590

Skype: xianfeng.cheng1

Email: chengxianfeng@speechocean.com; cxfxy0cxfxy0@gmail.com

Website: www.speechocean.com

Top

ROCme!: nouveau logiciel gratuit pour l'enregistrement et la gestion de corpus audio.

Le logiciel ROCme! permet une gestion rationalisée, autonome et dématérialisée de l’enregistrement de corpus lus.

Caractéristiques clés :
- gratuit
- compatible Windows et Mac
- interface paramétrable pour le recueil de métadonnées sur les locuteurs
- le locuteur fait défiler les phrases à l'écran et les enregistre de façon autonome
- format audio paramétrable

Téléchargeable à cette adresse :
www.ddl.ish-lyon.cnrs.fr/rocme

Top

VocalTractLab 2.0 : A tool for articulatory speech synthesis

It is my pleasure to announce the release of the new major version 2.0 of VocalTractLab. VocalTractLab is an articulatory speech synthesizer and a tool to visualize and explore the mechanism of speech production with regard to articulation, acoustics, and control. It is available from http://www.vocaltractlab.de/index.php?page=vocaltractlab-download .
Compared to version 1.0, the new version brings many improvements in terms of the implemented models of the vocal tract, the vocal folds, the acoustic simulation, and articulatory control, as well as in terms of the user interface. Most importantly, the new version comes together with a manual.

If you like, give it a try. Reports on bugs and any other feedback are welcome.

Peter Birkholz

Top

    The release 1.2.0 of the Bob
      signal-processing and machine learning toolbox is available .
    Bob provides both efficient implementations of several machine     learning algorithms as well as a framework to help researchers to     publish reproducible research.
   
   

   
    The previous release of Bob was providing:
    * image, video and audio IO
      interfaces such as jpg, avi, wav, 
    * database
      accessors such as FRGC, Labelled Face in the Wild, and many     others,
    *       image processing: Local Binary Patterns (LBPs), Gabor Jets,     SIFT,
    * machines
      and trainers such as Support Vector Machines (SVMs), k-Means,     Gaussian Mixture Models (GMMs), Inter-Session Variability modeling     (ISV), Joint Factor Analysis (JFA), Probabilistic Linear     Discriminant Analysis (PLDA), Bayesian intra/extra (personal)     classifier,
   
    The new release of Bob has brought the following features and/or     improvements, such as:
    * Unified implementation of Local Binary Patterns (LBPs),
    * Histograms of Oriented Gradients (HOG) implementation,
    * Total variability (i-vector) implementation,
    * Conjugate gradient based-implementation for logistic regression,
    * Improved multi-layer perceptrons implementation (Back-propagation     can now be easily used in combination with any optimizer -- i.e     L-BFGS),
    * Pseudo-inverse-based method for Linear Discriminant Analysis,
    * Covariance-based method for Principal Component Analysis,
    * Whitening and within-class covariance normalization techniques,
    * Module for object detection and keypoint localization     (bob.visioner),
    * Module for       audio processing including feature extraction such as LFCC and     MFCC,
    * Improved extensions (satellite packages), that now support both     Python and C++ code, within an easy to use framework,
    * Improved documentation and add new tutorials,
    * Support for Intel's MKL (in addition to ATLAS),
    * Extend supported platforms (Arch Linux).
   
    This release represents a major milestone in Bob with plenty of     functionality improvements (>640       commits in total) and plenty of bug
      fixes.
    • Sources and       Documentation
    • Binary packages:
    •     Ubuntu: 10.04, 12.04, 12.10 and 13.04
    • For     Mac OSX: works with 10.6 (Snow Leopard), 10.7 (Lion) and 10.8     (Mountain Lion)
   
    For instructions on how to install pre-packaged version on Ubuntu or     OSX, consult our quick       installation instructions  (N.B. OS X macport has not yet been     upgraded. This will be done very soon. cf. https://trac.macports.org/ticket/39831 ).
   
   
    Best regards,
    Elie Khoury (on Behalf of the Biometric
      Group at Idiap lead by Sebastien
      Marcel)
   
     
     ---    

-- ------------------- Dr. Elie Khoury Post Doctorant Biometric Person Recognition Group IDIAP Research Institute (Switzerland) Tel : +41 27 721 77 23

Top

Release of the version 2 of FASST (Flexible Audio Source Separation Toolbox). http://bass-db.gforge.inria.fr/fasst/ This toolbox is intended to speed up the conception and to automate the implementation of new model-based audio source separation algorithms. It has the following additions compared to version 1: * Core in C++ * User scripts in MATLAB or python * Speedup * Multichannel audio input We provide 2 examples: 1. two-channel instantaneous NMF 2. real-world speech enhancement (2nd CHiME Challenge, Track 1)

Top