Speech synthesis using generalized cepstrum
Takao Kobayashi, Satoshi Imai · Electronics and Communications in Japan (Part I Communications) · 1982
Abstract As a speech analysis method using the generalized cepstrum there is a method based on a generalized logarithmic function instead of a logarithmic function as an evaluation scale. This paper presents a speech synthesis system using the generalized cepstrum, which is defined as the Fourier coefficients of the generalized logarithmic spectrum. The direct approximated digital filter using the Padé approximation, which is the inverse function of the generalized logarithmic function, is used as a synthesis filter. Since this filter not only gives a good approximation of the desired generalized logarithmic spectrum by a generalized cepstrum but also since the values of generalized cepstra correspond to the filter coefficients as they are, it is easy to determine the synthesis filter coefficients. Since the synthesis system in this paper is fully compatible with the one based on the conventional cepstrum method, it is possible to synthesize speech by the generalized cepstrum without changing the speech synthesis system based on the conventional cepstrum method. Also, using this synthesis system, it is possible to synthesize speech by the conventional cepstrum method as a special case of this method. With proper choice of the parameters of the generalized logarithmic function, the synthesized speech by this method is of higher quality than that by the conventional cepstrum method.