|
| The CNRS | LIMSI Directory | |||
WebmasterSpoken Language Processing Group (TLP)
TLP Group - Presentation
The Spoken Language Processing group carries out research aimed at understanding the human speech communication processes and developing models for use in automatic processing of speech. This research is by nature interdisciplinary, drawing upon expertise in signal processing, acoustic-phonetics, phonology, semantics, statistics and computer science. The group's research activities are validated by developing systems for automatic processing of spoken language such as speech recognition, language identification, multimodal characterization of speakers and their affective state, named-entity extraction and question- answering, spoken dialog, multimodal indexation of audio and video documents, and machine translation of both spoken and written language.
With the aim of extracting and structuring information in audio documents, the group develops models and algorithms that use diverse sources of information to carry out a global decoding of the signal, that can be applied to identify the speaker, the language being spoken if it is not known a priori, the affect, to transcribe the speech or translate it, or identify specific entities.
Speech recognition is the process of transcribing the speech signal into text. Depending upon the targeted use, the transcription can be completed with punctuation, with paralinguistic information such as hesitations, laughter or breath noises. Research on speech recognition relies on supporting research in acoustic-phonetic modeling, lexical modeling and language modeling (a problem also addressed for machine translation), which are undertaken in a multilingual context (18 languages).
Statistical machine translation is an intensive area of research for the group today with the development of novel language and translation models as well as novel decoding strategies. This research area is closely related to the development of machine learning tools with two major achievements: the Wapita open source software for linear chain CRFs, and the development of new tools for neural network language model training.
Affective and social dimension detection are being applied to both human-machine interaction with robots and in the analysis of audiovisual documents such as call center data. The main research subjects in this area are speaker and emotion identification in human-robot interaction, emotion detection in client/agent interaction, emotion detection based on acoustic, visual and physiological cues for assistive robotics, and ,ultimodal detection of the anxiety applied to therapeutic serious games.
Robust analysis methods for the spoken language are being developed in the framework of open domain information retrieval with applications to language understanding for dialog systems, to named- entity recognition, and to interactive question answering systems supporting both spoken and written languages.
As of December 2011, the group has 43 members -- 12 permanent CNRS, 6 research associates, 11 postdocs, 2 contractual research staff, and 12 doctoral students. In addition to its research activities, the group is responsible for several graduate level speech processing courses, principally at the University of Paris-Sud. In 2010 and 2011 the members of the group published 147 articles (21 in journals, 28 chapters in books, and 97 reviewed conference papers).
| Activities | Projects | Publications | People |
Last modified: Saturday,01-September-12 17:40:32 CEST