Labs partners


LIUM (coordinator)

  • LIUM is the Computer Science Laboratory of the Université du Maine (Le Mans, France). The LIUM’s competence concerned by the ASH project is the automatic speech processing. This competence is bringing by the LST team (LST: Language and Speech Technology).
  • Participating to the French ESTER evaluation campaign and to the 3rd ASR evaluation campaign of the European TC-STAR project, LIUM has developed efficiently entire automatic transcription and audio/speaker segmentation systems for French, English and Spanish languages. These systems represent a very good base to work on the ASH project (some parts of them are available on the LIUM website under open source license).
  • The automatic transcription system is based on the CMU Sphinx s3.x decoder. Recently, some researchers of the LIUM Speech Project have been integrated into the official CMU Sphinx Project Developer Team. Moreover, the LIUM is experienced about data fusion: in particular, the LIUM has worked for years on audio and video data fusion applied on speech recognition.


  • The LIA has a strong experience in the development of speech processing tools. In particular, a fully home-made speech recognition system, Speeral, was developed at LIA. Speeral has been evaluated during the ESTER broadcast news transcription evaluation campaign where it has obtained good results. It is used by several laboratories and companies (Voice Insight, Xtensive Technologies….).
  • LIA has been very active over the last few years in various aspects of speech processing : acoustic segmentation, speaker diarization and tracking, speech recognition under various constraints (embedded systems, real-time systems, low resourced languages…), speaker adaptation, dialogue systems, language and semantic modelling, etc.
  • LIA has also been active in various corpus based evaluation programs including all tasks evaluated in ESTER campaign (speech transcription, segmentation, speaker tracking and named entities extraction), MEDIA (understanding in spoken dialogue), EQUER (question answering), NIST (speaker recognition, Rich Transcription) and TREC (information retriveal).


  • The Metiss team at Irisa is dedicated to advanced speech and audio processing, with a strong expertise in source separation, speech and speaker recognition and, in relation with other teams at Irisa, multimedia modeling.
  • In the field of speech recognition, Guillaume Gravier, scientific leader of the current project for Irisa, has been the main developer of the Sirocco speech recognition system which is now at the heart of the Irisa spoken document indexing platform.
  • This platform was evaluated during the ESTER evaluation campaign in 2005 and has now reached state-of-the-art performance. Recent work related to the current project include an algorithm to introduce landmarks in a Viterbi-based decoder (Moraru et al. JEP 2006; Gravier et al. NOLISP 2007) as well as combining linguistic knowledge in ASR systems (Huet et al., TSD 2006).
  • Guillaume Gravier has also been active in the field of multimedia modeling since 2001 and has recently study audiovisual integration in HMM (Kijak et al. Multimedia Tools and Applications, 2005) and segment models (Delakis et al. ICME 2006) approaches to video analysis.



  • Coordinator: Yannick ESTEVE, Professor (LIUM)
  • Paul DELEGLISE, Professor (LIUM)
  • Guillaume Gravier, Research fellow, HDR (IRISA)
  • Georges LINARES, Assistant Professor, HDR (LIA)
  • Driss MATROUF, Assistant Professor, HDR (LIA)
  • Sylvain MEIGNIER, Assistant Professor (LIUM)
  • Pascal NOCERA, Assistant Professor (LIA)

Non-permanent (positions funded by the ANR)

  • Fethi Bougares, PhD student, directed by Paul DELEGLISE (LIUM), Yannick ESTEVE (LIUM) and Georges LINARES (LIA). From october 2009 to september 2012;
  • Benjamin Lecouteux, Post-doc, LIA. From october 2009 to september 2010.
people.txt · Last modified: 12/09/2010 19:09 by Yannick Estève