Second International Conference on Spoken Language Processing (ICSLP'92)

Banff, Alberta, Canada
October 13-16, 1992

Improved Connected Digit Recognition Using Spectral Variation Functions

F. Brugnara (1), Renato De Mori (2), D. Giuliani (1), Maurizio Omologo (1)

(1) Istituto per la Ricerca Scientifica e Tecnologica, Povo di Trento, Italy
(2) School of Computer Science, McGill University, Montreal, Quebec, Canada

The use of Spectral Variation Functions (SVFs) for Automatic Speech Recognition (ASR) is discussed. Two types of SVFs are compared and results for different SVF orders are presented. A 20% string error-rate reduction has been obtained on the 10 kHz TI/NIST connected digit corpus, by adding a fourth order SVF to a set of acoustic features made of S Mel scaled Cepstral Coefficients(MCC), their time derivatives, the normalized signal energy and its time derivative.

Full Paper

Bibliographic reference.  Brugnara, F. / Mori, Renato De / Giuliani, D. / Omologo, Maurizio (1992): "Improved connected digit recognition using spectral variation functions", In ICSLP-1992, 627-630.