Exploring GMM-derived Features for Unsupervised Adaptation of Deep Neural Network Acoustic Models

Abstract : In this paper we investigate GMM-derived features recently introduced for adaptation of context-dependent deep neural network HMM (CD-DNN-HMM) acoustic models. We present an initial attempt of improving the previously proposed adaptation algorithm by applying lattice scores and by using con dence measures in the traditional max- imum a posteriori adaptation (MAP) adaptation algorithm. Modi ed MAP adaptation is performed for the auxiliary GMM model used in a speaker adaptation procedure for a DNN. In addition we introduce two approaches - data augmentation and data selection, for improving the regularization in MAP adaptation for DNN. Experimental results on the Wall Street Journal (WSJ0) corpus show that the proposed adaptation technique can provide, on average, up to 9:9% relative word error rate (WER) reduction under an unsupervised adaptation setup, compared to speaker independent DNN-HMM systems built on conventional features.
Type de document :
Communication dans un congrès
18th International Conference on Speech and Computer, 2016, Budapest, Hungary. 2016, Speech and Computer 18th International Conference, SPECOM 2016, Budapest, Hungary, August 23-27, 2016, Proceedings
Liste complète des métadonnées

https://hal.archives-ouvertes.fr/hal-01433184
Contributeur : Sylvain Meignier <>
Soumis le : lundi 19 novembre 2018 - 09:35:10
Dernière modification le : jeudi 29 novembre 2018 - 11:18:42
Document(s) archivé(s) le : mercredi 20 février 2019 - 13:13:47

Fichier

ExploringGMMDfeatures_SPECOM20...
Fichiers produits par l'(les) auteur(s)

Identifiants

  • HAL Id : hal-01433184, version 1

Collections

Citation

Natalia Tomashenko, Yuri Khokhlov, Anthony Larcher, Yannick Estève. Exploring GMM-derived Features for Unsupervised Adaptation of Deep Neural Network Acoustic Models. 18th International Conference on Speech and Computer, 2016, Budapest, Hungary. 2016, Speech and Computer 18th International Conference, SPECOM 2016, Budapest, Hungary, August 23-27, 2016, Proceedings. 〈hal-01433184〉

Partager

Métriques

Consultations de la notice

99

Téléchargements de fichiers

19