Performance of a 200-Syllable Communication System

Abstract

A 200-syllable communication system has been built and tested that automatically analyzes, recognizes, and encodes speech at the transmitter. At the receiver, the digital code is used to synthesize speech from prerecorded speech elements. The result is highly intelligible speech transmitted at the extremely low bit rate of 8 bits/sec. This system is the result of several years of research on speech processing directed toward discovering the useful, invariant information-bearing portions of the acoustic signal. The analyzer processes speech spoken one syllable at a time and transmits an eight-bit code for each syllable. The features used to derive the code include intrasyllabic pauses, voiced-to-unvoiced sound transitions, rate of growth and decay, duration, and relative spectral information. The entire communication system has been completed and the memories have been charged to store the codes for 200 syllables. An overall performance evaluation has been made using four speakers over a ten-day period with a 24 000-voicing-test sample. The system responded to 91% of the voicings with an accuracy of 80%.