Presentation 1998/6/11
An Evaluation of Mel-LPC Analysis Method in Speech Recognition
Yoshihisa Nakatoh, Hiroshi Matsumoto,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) The linear prediction analysis on a mel- or bark-frequency scale proposed by Strube is expected to be effective in speech recognition as the MFCC or PLP analysis because of their auditory-like frequency resolution. However, this method has been rarely used speech recognition applications due to relatively high computational load compared to the standard LPC analysis. This paper proposes a simple and efficient time-domain technique (Mel-LPC analysis) to estimate warped predictors. This analysis is accomplished with about two-fold increase in computation over the standard LPC analysis. The recognition performance of mel-cepstral parameters obtained by the Mel-LPC analysis is compared with that of the conventional LPC mel-cepstra through speaker independent phoneme recognition. The results show that the Mel-LPC cepstrum leads to the improvement from 64.8% for the standard LPC mel-cepstrum to 73.4% in phoneme recognition accuracy, and the improvement from 92.7% for the standard LPC mel-cepstrum to 96.0% in word recognition accuracy (520 vocabulary word).
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Mel-LPC analysis / Speech recognition / Mel-cepstrum / Frequency warping
Paper # SP98-22
Date of Issue

Conference Information
Committee SP
Conference Date 1998/6/11(1days)
Place (in Japanese) (See Japanese page)
Place (in English)
Topics (in Japanese) (See Japanese page)
Topics (in English)
Chair
Vice Chair
Secretary
Assistant

Paper Information
Registration To Speech (SP)
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) An Evaluation of Mel-LPC Analysis Method in Speech Recognition
Sub Title (in English)
Keyword(1) Mel-LPC analysis
Keyword(2) Speech recognition
Keyword(3) Mel-cepstrum
Keyword(4) Frequency warping
1st Author's Name Yoshihisa Nakatoh
1st Author's Affiliation Multimedia Development Center, Matsushita Electric Industrial Co., Ltd.()
2nd Author's Name Hiroshi Matsumoto
2nd Author's Affiliation Faculty of Engineering, Shinshu University
Date 1998/6/11
Paper # SP98-22
Volume (vol) vol.98
Number (no) 105
Page pp.pp.-
#Pages 8
Date of Issue