But neural network features for spontaneous Vietnamese in BABEL

Author

Karafiat, Martin ; Grezl, Frantisek ; Hannemann, Mirko ; Cernocky, Jan Honza

Author_Institution

Speech@FIT, Brno Univ. of Technol., Brno, Czech Republic

fYear

2014

fDate

4-9 May 2014

Firstpage

5622

Lastpage

5626

Abstract

This paper presents our work on speech recognition of Vietnamese spontaneous telephone conversations. It focuses on feature extraction by Stacked Bottle-Neck neural networks: several improvements such as semi-supervised training on untranscribed data, increasing of precision of state targets, and CMLLR adaptations were investigated. We have also tested speaker adaptive training of this architecture and significant gain was found. The results are reported on BABEL Vietnamese data.

Keywords

feature extraction; neural nets; regression analysis; speech recognition; BABEL Vietnamese data; CMLLR adaptations; Vietnamese spontaneous telephone conversations; constrained maximum likelihood linear regression; feature extraction; semisupervised training; speaker adaptive training; speech recognition; stacked bottleneck neural networks; untranscribed data; Artificial neural networks; Feature extraction; Hidden Markov models; Speech; Speech recognition; Training; adaptation of neural networks; bottleneck neural networks; discriminative training; region-dependent transforms; speech recognition;

fLanguage

English

Publisher

ieee

Conference_Titel

Acoustics, Speech and Signal Processing (ICASSP), 2014 IEEE International Conference on

Conference_Location

Florence

Type

conf

DOI

10.1109/ICASSP.2014.6854679

Filename

6854679