An investigation into VTLN for improved transcription of Czech broadcast programs

Author

Cerva, Petr ; Palecek, Karel ; Silovsky, Jan ; Nouza, Jan

Author_Institution

Inst. of Inf. Technol. & Electron., Tech. Univ. of Liberec, Liberec, Czech Republic

fYear

2011

Firstpage

201

Lastpage

204

Abstract

This paper deals with the Vocal Tract Length Normalization (VTLN) method. The aim is to investigate the best way how to utilize this technique for improving recognition accuracy of a LVCRS system that has been developed for broadcast program transcription at our lab in recent years. For this purpose, VTLN is evaluated experimentally in several configurations during testing as well as in speaker adaptive training scheme. In the former case, we employ VTLN as unsupervised for each testing utterance without the knowledge of transcription of adaptation data. Our results on different types of broadcast programs show that the resulting approach for VTLN reduces the Word Error Rate (WER) of our system significantly - by 7 % relatively.

Keywords

broadcasting; speaker recognition; unsupervised learning; Czech broadcast programs; LVCRS system; VTLN; broadcast program transcription; recognition accuracy; speaker adaptive training; testing utterance; unsupervised; vocal tract length normalization; word error rate; Acoustics; Hidden Markov models; Optimized production technology; Speech; Speech recognition; Testing; Training; Transcription of broadcast programs; Unsupervised speaker adaptation; Vocal tract length normalization;

fLanguage

English

Publisher

ieee

Conference_Titel

ELMAR, 2011 Proceedings

Conference_Location

Zadar

ISSN

1334-2630

Print_ISBN

978-1-61284-949-2

Electronic_ISBN

1334-2630

Type

conf

Filename

6044296