Overall Objectives
Scientific Foundations
Application Domains
New Results
Contracts and Grants with Industry
Other Grants and Activities

Section: Software

Keywords : speech modeling, speech recognition, broadcast news indexing, beam-search, Viterbi, HMM.

Irene: a speech recognition and transcription platform

Participant : Guillaume Gravier.

In collaboration with the computer science dept. at ENST, METISS has actively participated in the past years in the development of the freely available Sirocco large vocabulary speech recognition software  [58] . The Sirocco project started as an INRIA Concerted Research Action now works on the basis of voluntary contributions.

The Sirocco speech recognition software was then used as the heart of the transcription modules whithin a spoken document analysis platform called IRENE. In particular, it has been extensively used for research on ASR and NLP as well as for work on phonetic landmarks in statistical speech recognition.

This year has been dedicated to a major refactoring of the ASR part ot the system. On the one hand, the architecture of the Irene transcription system has been redesigned to be part of the multimedia indexing platform jointly developped by the Texmex , Metiss and Vista project-teams, thus enabling the large scale transcription of TV programs.

On the other hand, a new version of the Irene transcription system was developped in collaboration with the Texmex project-team, using a larger amount of training data. In particular, we have upgraded the language model (LM) of the ASR system in order to enable vocabulary adaptation on top of LM adaptation. Significant improvements were benchmarked in the framework of the Ester 2 evaluation campaign on the transcription radio shows in the French language.

Contact :


Logo Inria