Sites Inria

Version française

METISS Research team

Speech and sound data modeling and processing

  • Leader : Frédéric Bimbot
  • Research center(s) : CRI Rennes - Bretagne Atlantique
  • Field : Perception, Cognition, Interaction
  • Theme : Audio, Speech, and Language Processing
  • Partner(s) : Université Rennes 1,CNRS
  • Collaborator(s) : U. RENNES 1, CNRS, INRIA, INSA RENNES, CENTRALESUPELEC, ENS RENNES, INSTITUT MINES-TELECOM, UBS

Team presentation

The application fields of METISS are centered on sound signals and have three facets: speaker characterization (in particular for vocal identification), speaker and sound class pursuit for the indexing of sound recordings, and advanced treatment of sound signals (for example, source separation in the under determined case). Our scientific activity is grounded on applied mathematics, signal processing, probabilistic modeling, statistical estimation and decision theory. We use signal processing tools at the level of signal representation (adaptive representations), parametrization (spectral analysis) and decomposition (source separation). Probabilistic approaches come in at the level of acoustic modeling (distribution models) and classification (hypothesis tests and recognition). Our work also calls on decoding and pursuit algorithms such as the Viterbi algorithm and the matching pursuit. The main industrial sectors concerned are telecommunications, the Internet and the multimedia industry. These sectors may be extended to the fields of musical and audiovisual production, and educational software and games.

Research themes

  • Speaker characterization, identification and verification
  • Modeling, information detection and audio recording indexing
  • Source separation and advanced sound processing

International and industrial relations

  • Avignon Computer Science Department (LIA), ENST, Lyon II - DDL, EPFL,... : ELISA consortium (Annual participation in the NIST evaluations in speaker recognition and pursuit [1997-...]
  • INA,CS-Systèmes d'Information, Arts Vidéo Interactive and Mémodata, as well as teams from IRIT, CLIP-IMAG, INT, and LIP6: RNRT AGIR project (development of an audiovisual indexing system and search by contents) [1998-2001]
  • Ibermatica, BBVA, Oberthur, Thalès Communication, and departments of EPFL, IDIAP, University Carlos III, University of Surrey: projet BANCA (speaker verification for bank transactions) [1999-2002]
  • CP8 (ex-Bull): Fast, distributed speaker verification on smart card [1999-2001]
  • Thomson MultiMédia, IRCCyn, INA, SFRS: RNRT Domus Videum project (Generating audiovisual summaries for home multimedia platforms [2001-2004].

Keywords: Modeling Experimentation Signals Sound