First International Conference on Spoken Language Processing (ICSLP 90)

Kobe, Japan
November 18-22, 1990

Enhanced Parametric Representation Using Binarized Spectrum

Ki Chul Kim, Hyun Soo Yoon, Jung Wan Cho

Department of Computer Science & Center for Artificial Intelligence Research, Korea Advanced Institute of Science and Technology, Seoul, Korea

This paper describes an enhanced parametric representation for all-pole models of speech useful for speech recognition. The enhanced parametric representation, mcl-frequency peak coefficients, utilizes the concepts of spectral peak weighting and mel-scale integration in the frequency domain. The speaker-dependent syllable recognition results show that mcl-frequency peak coefficients with the Euclidian distance measure offers improved performance in comparison to LPC and mel-cepstrum with root power sum distance measure.

Full Paper

Bibliographic reference.  Kim, Ki Chul / Yoon, Hyun Soo / Cho, Jung Wan (1990): "Enhanced parametric representation using binarized spectrum", In ICSLP-1990, 201-204.