A Speech Predictive Encoding Communication System for Multichannel Telephony

This paper describes a speech predictive encoding communications (SPEC) system that is designed to accommodate the traffic of
N
PCM telephone trunks in the capacity of
N/2
PCM telephone channels. The system exploits the statistics of multichannel voice communication in a manner significantly different from the technique employed in the well-known time-assignment speech interpolation (TASI) system. Specifically, the SPEC system operates at the transmitter by removing redundant speech samples during talk bursts as well as during silence intervals. The receiver remembers the most recent sample transmitted through each channel. As long as a new sample is not transmitted, the receiver simply repeats the sample value stored in its memory. Each sample from each incoming trunk is examined every 125 μs to determine which samples must be transmitted and which are redundant. Because the SPEC system operates at this high speed and predicts redundant or missing samples at the receiver, it does not produce the speech clipping effects typical of TASI-like systems. Any voice-quality degradation due to increased call load and/or speech activity is manifested as an increase in noise power. The subjective result is a graceful degradation that is virtually imperceptible to the listener.