This paper proposes a new linear predictive coding (LPC) based on the wavelet functions. It has been shown in the literature that the wavelet functions form a complete set of orthonormal basis to decompose the time signals. One of the main advantages of using the wavelet functions is their optimal time-frequency distribution which provides good resolution in analyzing the excitations. In the excitation model, the excitation signal comprises of a set of compound pulses. Each compound pulse contains L equally-spaced pulses, in which the location of the compound pulse and the amplitudes of the pulse elements are determined by analysis-by-synthesis technique to minimize the error between the original speech and synthetic speech. A real speech is used to evaluate the performance of the new model. The parameters of the model are chosen to limit the coding rate to 8 kbps. The new scheme is proved to provide a class of excitations having better performance than multipulse LPC model presented in .
Bibliographic reference. Leung, S. H. / Wong, O. Y. / Lai, K. L. (1991): "Decomposition of the LPC excitation using wavelet functions", In EUROSPEECH-1991, 1327-1330.