LPC and CCF vocal tract models in speech synthesis

In the paper a new cepstral continued fraction zero-pole model for speech synthesis is presented. It is based on the cosine expansion of the logarithmic short time spectrum of the windowed speech signal and the synthesis is realized by adaptive approximative inverse cepstral transform using continued fraction digital filter structures. These structures were implemented on a digital signal processor. The spectrum modelling properties for this approach and those obtained by linear predictive coding are compared.