|
| The CNRS | LIMSI Directory | |||
WebmasterSpoken Language Processing Group (TLP)
TLP Group - Presentation
The research carried out in the Spoken Language Processing Group aims at understanding the speech communication processes and developing models for use in automatic speech processing. The problems we address concern modeling at multiple levels acoustic, lexical, syntactic and semantic as well as the communication process. This research area is inherently multidisciplinary, requiring competence in signal processing, acoustics, phonetics and phonology, linguistics and computer science. These research activities are validated by developing systems for automatic speech processing such as speech recognition, spoken language dialog, and the indexation of audio and video documents.
Our research in speech recognition focuses on the segmentation and transcription of continuous speech as well as speaker and language identification, with the purpose of automatically annotating and structurizing audio documents. Advances in speech recognition, which rely upon supporting research in acoustic-phonetic modeling, lexical modeling and language modeling, are undertaken in a multilingual context (English, French, German, Mandarin, Portuguese, Spanish, Arabic, etc.). The underlying approach aims to develop models and algorithms that jointly take into account the diverse sources of information so as to globally decode the audio signal.
Human-machine spoken dialog is a multi-faceted research area requiring us to model spontaneous speech, the communication process and to develop a dialog manager. This work has led to the realization of dialog systems for information access. The interaction can rely entirely on speech (via the telephone), or can be associated with other means of interaction, such as a touch screen (multimodal kiosk).
A recent research area of the group is content-based indexation of audiovisual documents for information retrieval and media watch applications. We are developing efficient indexation techniques for audio documents, which do not have the same characteristics as textual documents. The huge quantity of data being produced on a regular basis has led us to develop new decoding solutions and techniques for estimating the recognition models in order to be able to exploit this data without first needing to manually transcribe it. Vocapia Research develop state-of-start speech-to-text system based on our technology.
Three complementary activities support the main research areas: the design and production of corpora, the evaluation of models and systems, and technology transfer, in particular in the context of European projects.
As of October 2007, the group has 30 members -- 10 permanents CNRS, 5 research associates, 6 postdocs, 1 contractual research staff, and 8 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 XI. From 2005 to 2007 the members of the group published 132 articles (19 in journals, 3 chapters in books, and 110 reviewed conference papers).
| Activities | Projects | Publications | People |
Last modified: Saturday,20-November-10 20:25:11 CET