EUROSPEECH 2003 - INTERSPEECH 2003
8th European Conference on Speech Communication and Technology

Geneva, Switzerland
September 1-4, 2003

        

Multi-Source Training and Adaptation for Generic Speech Recognition

Fabrice Lefevre, Jean-Luc Gauvain, Lori Lamel

LIMSI-CNRS, France

In recent years there has been a considerable amount of work devoted to porting speech recognizers to new tasks. Recognition systems are usually tuned to a particular task and porting the system to a new task (or language) is both time-consuming and expensive. In this paper, issues in speech recognition portability are addressed and in particular the development of generic models for speech recognition. Multi-source training techniques aimed at enhancing the genericity of some wide domain models are investigated. We show that multi-source training and adaptation can reduce the performance gap between task-independent and task-dependent acoustic models, and for some tasks even out-perform task-dependent acoustic models. Ces dernieres annees, des efforts considerables ont ete faits pour faciliter le transfert des systemes de reconnaissance de la parole vers de nouvelles taches. Les systemes sont generalement optimises sur une tache particuliere et leur transfert vers une nouvelle tache est fastidieux et tres couteux en temps. Dans ce papier, nous nous interesserons au probleme du transfert des systemes de reconnaissance, en particuliers au travers du developpement de modeles generiques pour la reconnaissance de la parole. Des techniques d'apprentissage multi-source visant a augmenter le niveau de genericite de modeles a large domaine sont etudiees. Nous montrons que l'apprentissage et l'adaptation multi-sources peuvent permettre de reduire l'ecart de performance entre des modeles independants et dependants de la tache, et meme pour certaines taches de depasser les performances des modeles dependants de la tache.

Full Paper

Bibliographic reference.  Lefevre, Fabrice / Gauvain, Jean-Luc / Lamel, Lori (2003): "Multi-source training and adaptation for generic speech recognition", In EUROSPEECH-2003, 1877-1880.