High-quality digital speech at 4 kb/s

A speech coder based on a single-pulse excitation code-excited linear predictive coding (SPE-CELP) model of linear-predictive coding (LPC) is proposed. An algorithm for determining the time instants of pitch periods within a short interval of periodic speech, which results in a time sequence of marker points that indicate the beginning of the pitch periods in the analyzed speech interval, is described. The LPC excitation is generated by a stochastic codebook for nonperiodic speech and by a single pulse per pitch period for periodic speech. The proper alignment of the excitation pulse is efficiently computed using dynamic programming. It is concluded that, at overall bit rates of around 3 kb/s, the coder produces significantly better speech quality than LPC10E, though the synthesized speech still sounds slightly buzzy for certain speakers