Gwénolé LECORVÉ

PhD, Associate professor


Univ Rennes (ENSSAT)
  / IRISA (Team Expression)

6, rue de Kerampont
22300 Lannion, France

FirstName . LastName w/o accents [at] irisa [dot] fr
+33 (0)2 96 46 90 64


Short bio

My research interests are focused on speech and natural language processing.

2014-present Permanent member in the research team Expression at IRISA (Lannion, France)
2012-present Associate professor at ENSSAT/University of Rennes 1 (Lannion, France)
2012-2013 Permanent member in the research team Cordial at IRISA (Lannion, France)
2011-2012 Postdoctoral researcher at Idiap Research Institute (Martigny, Switzerland).
2007-2010 Ph.D. candidate in computer science in the multimedia group (Texmex) at IRISA/INRIA (INSA de Rennes, France).
2007
  • M.Sc. in image processing and artificial intelligence at the Institut National des Sciences Appliquées (INSA de Rennes, France)
  • M.Eng. in computer science at the Institut National des Sciences Appliquées (INSA de Rennes, France)
2004 B.Sc. in mathematics and computer science at the Université de Bretagne Sud (Lorient, France)

Teaching

Currently (ENSSAT)

  • Machine Learning (M.Eng. & M.Sc.)
  • Speech processing (M.Sc.)
  • Artificial Intelligence (M.Eng.)
  • Distributed algorithmics (M.Eng. & M.Sc.)

Past activities (ENSSAT, INSA Rennes)

  • Java, object-oriented programming (B.Sc & M.Eng.)
  • Unix and operating systems (M.Eng.)
  • Databases (B.Sc.)
  • Data structures (B.Sc.)
  • Symbolic data mining (M.Eng.)
  • Introduction to algorithmics and functional programming with Scheme (B.Sc.)

Projects

Research

Pronunciation modeling 2012-present

  • Grapheme-to-phoneme conversion
  • Pronunciation adaptation
    • Sponteneous and expressive speech
    • Voice-specific adaptation
  • Collaborators: Dr. Raheel Qader, Dr. Damien Lolive, Dr. Marie Tahon, Pr. Pascale Sébillot, Dr. Katarina Bartkova

Linguistic registers 2015-present

  • Characterization of linguistic register descriptors
  • Extraction of sequential patterns
  • Automatic collection of register-specific texts
  • Collaborators: Dr. Nicolas Béchet, Pr. Delphine Battistelli, Dr. Jonathan Chevelu, Ms. Inès Dabbebi, Ms. Jade Mekki, Mr. Benoît Fournier, Mr. Hugo Ayats.

Text-to-speech (TTS)2012-present

  • TTS International Blizzard Challenge
  • Embedding of speech units
  • Massive datasets
  • Collaborators: Mr. Antoine Perquin, Dr. Damien Lolive, Dr. Laurent Amsaleg

Disfluency modeling 2016-present

  • Automatic insertion of disfluencies
  • Interruption point prediction
  • Natural language generation
  • Collaborators: Dr. Raheel Qader, Dr. Damien Lolive, Pr. Pascale Sébillot

Speech recognition and language modeling 2007-2012

  • Unsupervised topic adaptation of an automatic speech recognition system (Ph.D. thesis)
  • Linguistic adaptation for automatic speech recognition
  • Language modeling using recurrent neural networks
  • Collaborators: Dr. Guillaume Gravier, Pr. Pascale Sébillot, Dr. Petr Motlicek, Dr. John Dines, Pr. Thomas Hain, Dr. Camille Guinaudeau, Dr. Stéphane Huet.

Ph.D. work: This work focused on the linguistic adaptation of an automatic speech recognition system to topics encountered in spoken documents. This work claims to avoid using any a priori knowledge about encountered topics and to integrate natural language processing techniques. To do so, I have proposed an unsupervised scheme to automatically retrieve topic-specific documents from the Internet based on the sole automatic transcription of thematically coherent broadcast. These Web-based corpora have been efficiently used to adapt an automatic speech recognition (ASR) system. On the one hand, I have proposed to consider the topic-based adaptation of a statistical language model as the sole re-estimation of probabilities on topic-specific words. In practice, by automatically deriving topic terminologies from Web corpora, I have highlighted that using a refined feature selection strategy within the MDI adaptation method can yield recognition accuracy improvements of an ASR system. On the other hand, I am currently working on the topic-specific adaptation of an ASR system vocabulary by characterizing new words to be added thanks to paradigmatic relations with other words and to the information about their usage in topic-specific texts.

Post-doctoral work: This work aimed at developing new mechanisms for language model and vocabulary adaptation according to domains of interest for speech recognition. Especially, I studied the efficiency of unsupervised approaches compared to supervised ones. Additionally, I were also investigating various other aspects of language modeling: new paradigms of models (neural network language models) and multilingual language modeling.

Project involvement

  • H2020 NADINE digital iNtegrAteD system for the socIal support of migraNts / and refugEes
    • European Commission
    • 2018-2021
  • Kaligo DYS Learning writing with dysgraphic and dyspraxic children and young adults
    • Pôle Images & Réseaux / Regional + EU funding
    • Local coordinator
    • 2018-2020
  • ANR TREMoLo: Language register transformation using linguistic pattern extraction
    • French National Research Agency
    • Coordinator
    • 2017-2021
  • PEPS TextToKids: Language register transformation using linguistic pattern extraction
    • CNRS (French National Center for Scientific Research)
    • Coordinator
    • 2018
  • ANR SynPaFlex: Flexible speech synthesis
    • French National Research Agency
    • Task leader on pronunciation adaptation
    • 2015-2019
  • TAO CSR: Task adaptation and optimization for continuous speech recognition.
    • Swiss Technology and Innovation Committee
    • 2011-2012

Publications

Thesis

  1. Gwénolé Lecorvé. Adaptation thématique non supervisée d'un système de reconnaissance automatique de la parole. PhD thesis, INSA de Rennes, 2010. AFCP Best PhD Award PDF

Book chapters (1)

  1. Stéphane Huet, Gwénolé Lecorvé, Guillaume Gravier, Pascale Sébillot. Multimodal Processing and Interaction: Audio, Video, Text. Petros Maragos, Alexandros Potamianos, Patrick Gros (eds.), Chapter Toward the Integration of Natural Language Processing and Automatic Speech Recognition: Using Morpho-syntax and pragmatics for transcription, Springer, 2008.

International journals (2)

  1. Guillaume Gravier, Camille Guinaudeau, Gwénole Lecorvé, Pascale Sébillot. Exploiting speech for automatic TV delinearization: from streams to cross-media semantic navigation, EURASIP Journal on Image and Video Processing, 2011. PDF
  2. Marie Tahon, Gwénolé Lecorvé, Damien Lolive. Can we Generate Emotional Pronunciations for Expressive Speech Synthesis?, IEEE Transactions on Affective Computing, 2018. To appear

International conferences (16)

  1. Gwénolé Lecorvé, Guillaume Gravier, Pascale Sébillot. An unsupervised Web-based topic language model adaptation method.  in Proc. of the IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP), 2008. PDF
  2. Gwénolé Lecorvé, Guillaume Gravier, Pascale Sébillot. On the use of Web resources and natural language processing techniques to improve automatic speech recognition systems.  in Proc. of the Conference on Language Resources and Evaluation (LREC), 2008. PDF
  3. Gwénolé Lecorvé, Guillaume Gravier, Pascale Sébillot. Constraint selection for topic-based MDI adaptation of language models  in Proc. of Annual Conference of the International Speech Communication Association (Interspeech), 2009. PDF
  4. Gwénolé Lecorvé, Guillaume Gravier, Pascale Sébillot. Automatically finding semantically consistent n-grams to add new words in LVCSR systems, in Proc. of the IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP), 2011. PDF
  5. Gwénolé Lecorvé, John Dines, Thomas Hain, Petr Motlicek. Supervised and unsupervised Web-based language model domain adaptation  in Proc. of Annual Conference of the International Speech Communication Association (Interspeech), 2012. PDF
  6. Gwénolé Lecorvé, Petr Motlicek. Conversion of recurrent neural network language models to weighted finite state transducers for automatic speech recognition  in Proc. of Annual Conference of the International Speech Communication Association (Interspeech), 2012. PDF
  7. David Imseng, Hervé Bourlard, Holger Caesar, Philip N. Garner, Gwénolé Lecorvé, Alexandre Nanchen. MediaParl: bilingual mixed language accented speech database in Proc. of IEEE Workshop on Spoken Language Technology (SLT), 2012. PDF
  8. Jonathan Chevelu, Gwénolé Lecorvé, Damien Lolive. ROOTS: a toolkit for easy, fast and consistent processing of large sequential annotated data collections in Proc. of the Conference on Language Resources and Evaluation (LREC), 2014. PDF
  9. Gwénolé Lecorvé, Damien Lolive. Adaptive statistical utterance phonetization for French in Proc. of the IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP), 2015. PDF
  10. Raheel Qader, Gwénolé Lecorvé, Damien Lolive, Pascale Sébillot. Probabilistic Speaker Pronunciation Adaptation for Spontaneous Speech Synthesis Using Linguistic Features. in Proc. of the International Conference on Statistical Language and Speech Processing (SLSP), Budapest, Hungary, 2015. PDF
  11. Marie Tahon, Raheel Qader, Gwénolé Lecorvé, Damien Lolive. Improving TTS with corpus-specific pronunciation adaptation, dans Proc. of the Annual Conference of the International Speech Communication Association (Interspeech), 2016. PDF
  12. Marie Tahon, Raheel Qader, Gwénolé Lecorvé, Damien Lolive. Optimal feature set and minimal training size for pronunciation adaptation in TTS, dans Proc. of the International Conference on Statistical Language and Speech Processing (SLSP), Pilsen, Czech Republic, 2016. PDF
  13. Raheel Qader, Gwénolé Lecorvé, Damien Lolive, Marie Tahon, Pascale Sébillot. Statistical Pronunciation Adaptation for Spontaneous Speech Synthesis. in Proc. of Text, Speech and Dialogue (TSD), Prague, Czech Republic, 2017. To appear
  14. Marie Tahon, Gwénolé Lecorvé, Damien Lolive, Raheel Qader. Perception of expressivity in TTS: linguistics, phonetics or prosody?, dans Proc. of the International Conference on Statistical Language and Speech Processing (SLSP), Le Mans, France, 2017. To appear
  15. Antoine Perquin, Gwénolé Lecorvé, Damien Lolive and Laurent Amsaleg. Phone-Level Embeddings for Unit Selection Speech Synthesis, dans Proc. of the International Conference on Statistical Language and Speech Processing (SLSP), Mons, Belgique, 2018. To appear
  16. Raheel Qader, Gwénolé Lecorvé, Damien Lolive and Pascale Sébillot. Disfluency Insertion for Spontaneous TTS: Formalization and Proof of Concept, dans Proc. of the International Conference on Statistical Language and Speech Processing (SLSP), Mons, Belgique, 2018. To appear

International workshops (2)

  1. Pierre Alain, Jonathan Chevelu, David Guennec, Gwénolé Lecorvé, Damien Lolive. The IRISA Text-To-Speech System for the Blizzard Challenge 2015. In Proc. of the Blizzard Challenge 2015 Workshop, 2015, Berlin, Germany. PDF
  2. Pierre Alain, Jonathan Chevelu, David Guennec, Gwénolé Lecorvé, Damien Lolive. The IRISA Text-To-Speech System for the Blizzard Challenge 2016. In Proc. of the Blizzard Challenge 2016 Workshop, 2016, Sunnyvale, USA. PDF
  3. Pierre Alain, nelly Barbot, Jonathan Chevelu, Gwénolé Lecorvé, Damien Lolive, Claude Simon, Marie Tahon. The IRISA Text-To-Speech System for the Blizzard Challenge 2017. In Proc. of the Blizzard Challenge 2017 Workshop, 2017, Stockholm, Sweden. To appear

French-speaking conferences (9)

  1. Gwénolé Lecorvé, Guillaume Gravier, Pascale Sébillot. Vers une adaptation thématique non supervisée de modèles de langage : utilisation d'Internet comme un corpus ouvert  in Actes des Journées d'Études sur la Parole (JEP), 2008. PDF
  2. Gwénolé Lecorvé, Guillaume Gravier, Pascale Sébillot. L'adaptation thématique d'un modèle de langue fait-elle apparaître des mots thématiques ?  in Actes des Journées d'Études sur la Parole (JEP), 2010. PDF
  3. Gwénolé Lecorvé, John Dines, Thomas Hain, Petr Motlicek. Impact du degré de supervision sur l'adaptation à un domaine d'un modèle de langage à partir du Web in Actes des Journées d'Études sur la Parole (JEP), 2012. PDF
  4. Jonathan Chevelu, Gwénolé Lecorvé, Damien Lolive. ROOTS : un outil pour manipuler facilement, efficacement et avec cohérence des corpus annotés de séquences. Actes des Journées d'Études sur la Parole (JEP), 2014. PDF
  5. Gwénolé Lecorvé, Damien Lolive. Phonétisation statistique adaptable d'énoncés pour le français. Actes des Journées d'Études sur la Parole (JEP), 2016. PDF
  6. Raheel Qader, Gwénolé Lecorvé, Damien Lolive, Pascale Sébillot. Adaptation de la prononciation pour la synthèse de la parole spontanée en utilisant des informations linguistiques. Actes des Journées d'Études sur la Parole (JEP), 2016. PDF
  7. Raheel Qader, Gwénolé Lecorvé, Damien Lolive, Pascale Sébillot. Ajout automatique de disfluences pour la synthèse de la parole spontanée : formalisation et preuve de concept. Actes des Traitement Automatique du Langage Naturel (TALN), long paper, 2017. Best Paper Award PDF
  8. Gwénolé Lecorvé, Hugo Ayats, Benoît Fournier, Jade Mekki, Jonathan Chevelu, Delphine Battistelli, Nicolas Béchet. Construction conjointe d'un corpus et d'un classifieur pour les registres de langue en français. Actes des Traitement Automatique du Langage Naturel (TALN), long paper, 2018. To appear
  9. Jade Mekki, Delphine Battistelli, Gwénolé Lecorvé, Nicolas Béchet. Identification de descripteurs pour la caractérisation de registrest. Actes des Rencontres des Jeunes Chercheurs de la conférence CORIA-TALN, 2018. To appear

Software

rnn2wfst

Conversion of recurrent neural network language models to weighted finite state transducers. Cite as follows:
@inproceedings{lecorve2012conversion, title={Conversion of recurrent neural network language models to weighted finite state transducers for automatic speech recognition}, author={Lecorv{\'e}, Gw{\'e}nol{\'e} and Motlicek, Petr}, booktitle={Proceedings of the Annual Conference of the International Speech Communication Association (Interspeech)}, year={2012}, location={Portland, Oregon, USA} }

> More on GitHub

Roots

Open source toolkit dedicated to annotated sequential data generation, management and processing. Cite as follows:
@inproceedings{chevelu2014roots, title={ROOTS: a toolkit for easy, fast and consistent processing of large sequential annotated data collections.}, author={Chevelu, Jonathan and Lecorv{\'e}, Gw{\'e}nol{\'e} and Lolive, Damien}, booktitle={Proceedings of the Language Resources and Evaluation Conference (LREC)}, pages={619--626}, year={2014}, location={Reykjavik, Iceland} }

> More on Bitbucket
> More on INRIA GForge

IRISA Text Normalizer

Scripts (Perl) to tokenize and normalize texts (French and English supported yet).

@misc{lecorve2017normalizer, title={The IRISA Text Normalizer}, author={Lecorv{\'e}, Gw{\'e}nol{\'e}}, howpublished={\url{https://github.com/glecorve/irisa-text-normalizer}}, year={2017} }
> More on GitHub