Skip to Main content Skip to Navigation
Theses

Détection de thème et adaptation des modèles de langage pour la reconnaissance automatique de la parole

Abstract : One way to improve performance of Automatic Speech Recognition (ASR) systems, consists in adapting language models to the topic treated in data. In this thesis, we propose a new vocabulary selection principle, resulting in a slight improvement of the performance. We also present anew topic identification method, WSIM, based on the similarity between words and topics, reaching performance similar to state of the art one. We have studied the evolution of the performance when methods are combined, reaching more than 93% correct topic identification. In the framework of ASR, adapting language model to the topic results in a large improvement of the perplexity.
File URL :
http://docnum.univ-lorraine.fr/prive/SCD_T_2003_0003_BRUN.pdf
Complete list of metadatas

https://hal.univ-lorraine.fr/tel-01746767
Contributor : Thèses Ul <>
Submitted on : Thursday, March 29, 2018 - 10:44:28 AM
Last modification on : Tuesday, April 24, 2018 - 1:30:18 PM

Identifiers

  • HAL Id : tel-01746767, version 1

Collections

Citation

Armelle Brun. Détection de thème et adaptation des modèles de langage pour la reconnaissance automatique de la parole. Autre [cs.OH]. Université Henri Poincaré - Nancy 1, 2003. Français. ⟨NNT : 2003NAN10003⟩. ⟨tel-01746767⟩

Share

Metrics

Record views

22