Exploitation d'une marge de tolérance de classification pour améliorer l'apprentissage de modèles acoustiques de classes en reconnaissance de la parole

Denis Jouvet 1 Arseniy Gorin 1 Nicolas Vinuesa 1
1 PAROLE - Analysis, perception and recognition of speech
Inria Nancy - Grand Est, LORIA - NLPKD - Department of Natural Language Processing & Knowledge Discovery
Abstract : This paper presents the introduction of a classification tolerance margin in the classification of the training data for building class-based acoustic models for automatic speech transcription. Indeed, although automatic classification of speech data makes it possible to go beyond the traditional male / female partition, the number of usable classes is actually limited by the reliability of the associated acoustic models which, unfortunately, decreases when the number of classes increases. The reported experiments show that using a tolerance margin in the classification process increases the amount of training data associated to each class, and consequently increases the reliability of the acoustic models of the classes. The performance evaluation carried on the ESTER2 data have shown that it is possible with the proposed approach to build class-based acoustic models that lead to better speech recognition performance than with the usual gender-based acoustic models.
Complete list of metadatas

https://hal.inria.fr/hal-00753394
Contributor : Denis Jouvet <>
Submitted on : Monday, November 19, 2012 - 11:06:49 AM
Last modification on : Tuesday, December 18, 2018 - 4:38:02 PM

Identifiers

  • HAL Id : hal-00753394, version 1

Collections

Citation

Denis Jouvet, Arseniy Gorin, Nicolas Vinuesa. Exploitation d'une marge de tolérance de classification pour améliorer l'apprentissage de modèles acoustiques de classes en reconnaissance de la parole. JEP-TALN-RECITAL 2012, Jun 2012, Grenoble, France. pp.763-770. ⟨hal-00753394⟩

Share

Metrics

Record views

451