Composite Cepstral Model for Speech Synthesis

Z. Smékal and M. Vondra (Czech Republic)


Speech Synthesis, Vocal Tract Model, Digital Signal Processors.


Speech is synthesized via exciting the vocal tract model by pulmonary air stream, which can be modulated by vocal cords. The vocal tract model is a time-variant digital filter designed to best simulate the transfer properties of the human vocal tract in the course of utterance. The paper describes the realization algorithm of the cepstral vocal tract model, which is optimized from the viewpoint of implementation on a fixed-point digital signal processor. Cepstral coefficients for speech synthesis are in fact coefficients of type FIR digital filter, which are introduced in the first and second canonic forms of type IIR digital filter instead of the delay blocks. The orders of the FIR and IIR filters depend on approximation accuracy and on sampling frequency.

Important Links:

Go Back