A frequency warping approach to speaker normalization

Author

Lee, Li ; Rose, Richard

Author_Institution

MIT, Cambridge, MA, USA

Volume

6

Issue

1

fYear

1998

fDate

1/1/1998 12:00:00 AM

Firstpage

49

Lastpage

60

Abstract

In an effort to reduce the degradation in speech recognition performance caused by variation in vocal tract shape among speakers, a frequency warping approach to speaker normalization is investigated. A set of low complexity, maximum likelihood based frequency warping procedures have been applied to speaker normalization for a telephone based connected digit recognition task. This paper presents an efficient means for estimating a linear frequency warping factor and a simple mechanism for implementing frequency warping by modifying the filterbank in mel-frequency cepstrum feature analysis. An experimental study comparing these techniques to other well-known techniques for reducing variability is described. The results have shown that frequency warping is consistently able to reduce word error rate by 20% even for very short utterances

Keywords

band-pass filters; cepstral analysis; filtering theory; frequency estimation; hidden Markov models; maximum likelihood estimation; speech processing; speech recognition; continuous speech recognition; experiment; filterbank; hidden Markov model; linear frequency warping factor estimation; low complexity frequency warping; maximum likelihood based frequency warping; mel-frequency cepstrum feature analysis; speaker normalization; speech recognition performance; telephone based connected digit recognition; very short utterances; vocal tract shape; word error rate reduction; Cepstral analysis; Cepstrum; Degradation; Error analysis; Filter bank; Frequency estimation; Maximum likelihood estimation; Shape; Speech recognition; Telephony;

fLanguage

English

Journal_Title

Speech and Audio Processing, IEEE Transactions on

Publisher

ieee

ISSN

1063-6676

Type

jour

DOI

10.1109/89.650310

Filename

650310