Vers une adaptation autonome des modèles acoustiques multilingues pour le traitement automatique de la parole

Abstract : Automatic speech recognition technologies are now integrated into many systems. The performance of speech recognition systems for non-native speakers, however, continues to suffer high error rates, due to the difference between native and non-speech models trained. The making of recordings in large quantities of non-native speech is typically a very difficult and impractical to represent all the origins of the speakers. This thesis focuses on improving multilingual acoustic models for automatic phonetic transcription of speech such as “multilingual meeting”. There are several challenges in “multilingual meeting” speech: 1) there can be a conversation between native and non native speakers ; 2) there is not only one spoken language but several languages spoken by speakers from different origins ; 3) it is difficult to collect sufficient data to bootstrapping transcription systems. To meet these challenges, we propose a process of adaptation of multilingual acoustic models is called "autonomous adaptation". In autonomous adaptation, we studied several approaches for adapting multilingual acoustic models in unsupervised way (spoken languages and the origins of the speakers are not known in advance) and no additional data is used during the adaptation process. The approaches studied are decomposed into two modules. The first module called "the language observer" is to recover the linguistic information (spoken languages and the origins of the speakers) of the segments to be decoded. The second module is to adapt the multilingual acoustic model based on knowledge provided by the language observer. To evaluate the usefulness of autonomous adaptation of multilingual acoustic model, we use the test data, which are extracted from multilingual meeting corpus, containing the native and nonnative speech of three languages: English (EN), French (FR) and Vietnamese (VN). According to the experiment results, the autonomous adaptation shows promising results for non native speech but very slightly degrade performance on native speech. To improve the overall performance of transcription systems for all native and non native speech, we study several approaches for detecting non native speech and propose such a detector cascading with our self-adaptation process (autonomous adaptation). The results thus are the best among all experiments done on our corpus of multilingual meetings.
Document type :
Theses
Complete list of metadatas

Cited literature [62 references]  Display  Hide  Download

https://tel.archives-ouvertes.fr/tel-00685204
Contributor : Abes Star <>
Submitted on : Wednesday, April 4, 2012 - 2:55:33 PM
Last modification on : Thursday, October 11, 2018 - 8:48:01 AM
Long-term archiving on : Thursday, July 5, 2012 - 2:32:56 AM

File

20337_SAM_2011_archivage_1_.pd...
Version validated by the jury (STAR)

Identifiers

  • HAL Id : tel-00685204, version 1

Citation

Sethserey Sam. Vers une adaptation autonome des modèles acoustiques multilingues pour le traitement automatique de la parole. Autre [cs.OH]. Université de Grenoble, 2011. Français. ⟨NNT : 2011GRENM017⟩. ⟨tel-00685204⟩

Share

Metrics

Record views

826

Files downloads

460