Publication Details

Towards Lower Error Rates in Phoneme Recognition

SCHWARZ, P.; MATĚJKA, P.; ČERNOCKÝ, J. Towards Lower Error Rates in Phoneme Recognition. Lecture Notes in Computer Science, 2004, vol. 2004, no. 3206, p. 465-472. ISSN: 0302-9743.
Czech title
Postup k nižší chybovosti fonémového rozpoznávače
Type
journal article
Language
English
Authors
Schwarz Petr, Ing., Ph.D. (DCGM)
Matějka Pavel, Ing., Ph.D.
Černocký Jan, prof. Dr. Ing. (DCGM)
URL
Keywords

phoneme recognition, traps, speech recognition, feature extraction

Abstract

We investigate techniques for acoustic modeling in automaticrecognition of context-independent phoneme strings from the TIMITdatabase. The baseline phoneme recognizer is based on TempoRAl Patterns(TRAP). This recognizer is simplified to shorten processing times andreduce computational requirements. More states per phoneme and bi-gramlanguage models are incorporated into the system and evaluated. Thequestion of insufficient amount of training data is discussed and thesystem is improved. All modifications lead to a faster system withabout 23.6% relative improvement over the baseline in phoneme errorrate.

Published
2004
Pages
465–472
Journal
Lecture Notes in Computer Science, vol. 2004, no. 3206, ISSN 0302-9743
BibTeX
@article{BUT45739,
  author="Petr {Schwarz} and Pavel {Matějka} and Jan {Černocký}",
  title="Towards  Lower Error Rates in Phoneme Recognition",
  journal="Lecture Notes in Computer Science",
  year="2004",
  volume="2004",
  number="3206",
  pages="465--472",
  issn="0302-9743",
  url="http://www.springerlink.com/index/KBY35VBXY16WHV56"
}
Back to top