Fourth European Conference on Speech Communication and Technology

Madrid, Spain
September 18-21, 1995

Effect of Rasta-Type Processing for Speech Recognition with Speaking-Rate Mismatches

Harald Singer (1), Kuldip K. Paliwal (2), Tomohiko Beppu (1), Yoshinori Sagisaka (1)

(1) ATR Interpreting Tel. Res. Labs., Soraku-gun, Kyoto, Japan
(2) School of Mic. Engineering, Griffith University, Brisbane, Australia

In the present paper, we investigate the use of "Rasta-type cepstral processing techniques", for speech recognition under mismatched speaking rate conditions. The acoustic models are trained on an isolated-word speech data base and and then tested on a continuous speech data base. The speaking rates in the two data bases are significantly different. Using high resolution phoneme-context dependent models, the high-pass cepstrum is shown to perform comparable for matched conditions and outperforms the other techniques.

Full Paper

Bibliographic reference.  Singer, Harald / Paliwal, Kuldip K. / Beppu, Tomohiko / Sagisaka, Yoshinori (1995): "Effect of rasta-type processing for speech recognition with speaking-rate mismatches", In EUROSPEECH-1995, 487-490.