Adaptation of Unknown Data to Already Trained Speech Recognition System

Czech title

Adaptace neznamých dat na natrénovaný systém

Type

conference paper

Language

English

Authors

Grézl František, Ing., Ph.D. (DCGM)

Keywords

data adaptation

Abstract

This work deals with different speech rate for phoneme recognition based on a Tandem system with TRAP-feature processing. The estimation of rate is based on measuring entropy at the output of the phoneme classifier. Preliminary results obtained on TIMIT database show that this technique may lead to estimation of correct speech rate for mismatched data. This work brings further insight on temporal trajectories and systems based on them.

Annotation

This work deals with different speech rate for phoneme recognition based on a Tandem system with TRAP-feature processing. The estimation of rate is based on measuring entropy at the output of the phoneme classifier. Preliminary results obtained on TIMIT database show that this technique may lead to estimation of correct speech rate for mismatched data. This work brings further insight on temporal trajectories and systems based on them.

Published

2005

Pages

4

Proceedings

Sborník prací konference a souteze Student EEICT 2005

Conference

STUDENT EEICT 2005, Brno, CZ

ISBN

80-214-2890-2

Publisher

Faculty of Information Technology BUT

Place

Brno

BibTeX

@inproceedings{BUT22410,
  author="František {Grézl}",
  title="Adaptation of Unknown Data to Already Trained Speech Recognition System",
  booktitle="Sborník prací konference a souteze Student EEICT 2005",
  year="2005",
  pages="4",
  publisher="Faculty of Information Technology BUT",
  address="Brno",
  isbn="80-214-2890-2"
}