First International Conference on Spoken Language Processing (ICSLP 90)

Kobe, Japan
November 18-22, 1990

Generalized Cepstral Analysis of Speech - Unified Approach to LPC and Cepstral Method

Keiichi Tokuda (1), Tahao Kobayashi (2), Satoshi Imai (2)

(1) Department of Electrical and Electronic Engineering, Tokyo Institute of Technology, Tokyo, Japan
(2) Research Laboratory of Precision Machinery and Electronics, Tokyo Institute of Technology, Yokohama, Japan

This paper describes a spectral estimation method based on the generalized cepstral representation. The model spectrum represented by the generalized cepstrum varies from the all-pole spectrum to that represented by the cepstrum according to the value of the parameter 7 in the range of [-1,0]. We apply the criterion used in the unbiased estimation of log spectrum to the spectral model. As a result, the proposed method includes linear prediction and the unbiased estimation of log spectrum as the special cases. To solve the non-linear minimization problem involved in the method, we give an iterative algorithm whose convergence is guaranteed. The stability of the model solution is guaranteed. We also show some results of natural speech analysis to discuss the optimal value of 7 in the sense of minimizing the prediction error.

Full Paper

Bibliographic reference.  Tokuda, Keiichi / Kobayashi, Tahao / Imai, Satoshi (1990): "Generalized cepstral analysis of speech - unified approach to LPC and cepstral method", In ICSLP-1990, 37-40.