The information rate of the pitch signal in speech

Abstract

In speech transmission system by means of schems analysis-synthesis (vocoders), naturalness of the reproduced speech is largely influenced by characteristics of the source signal. Among several quantities that may describe the source wave, the socalled pitch signal is of particular interest from a practical point of view. Precise data on the time change of the pitch signal for continuous messages are needed for the purpose of evaluating the minimum requirement for the channel transmitting the pitch signal. In this study, the time variation of the fundamental frequency was analized in detail, and the bandwidth necessary to transmit the "pitch signal" was determined based on the frequency modulation model. The extent of degradation of naturalness was evaluated by a psychometric method when a speech wave was put through an analysis-synthesis system, in which only the pitch signal was deliberately distorted. Based on these data, the information rate of the pitch signal required for high quality speech transmission systems has been estimated at approximately 750 bits/sec, or 50 cps in terms of the bandwidth for the pitch signal.