An Efficient Linear-Prediction Vocoder

The method of linear prediction has proved quite popular and successful for use in speech compression systems. 1-4 In this method, speech is modeled as the output of an all-pole filter H(z) that is excited by a sequence of pulses separated by the pitch period for voiced sounds, or pseudo-random noise for unvoiced sounds. These assumptions imply that within a frame of speech the output speech sequence is given by p s(n) = £ aks(n -- k) + un, where p is the number of modeled poles, un is the appropriate input excitation, and the ajt's are the coefficients characterizing the filter (linear prediction coefficients). Figure 1 illustrates the frequencydomain, as well as the equivalent time-domain, model of linear-prediction speech production. To account for the nonstationary character of the speech waveform, the parameters a k of the modeled filter are periodically updated during successive speech frames.* Generation of speech in this method requires a knowledge of the pitch, the filter * A frame is a segment of speech thought adequate to assume stationarity of the speech process. Typical frame lengths employed range from 10 to 30 ms. 1693