Interruption techniques for efficient speech transmission

[PART ONE, Investigations.] This thesis describes investigations into the use of interruption techniques for efficient speech transmission. Bit-rate compressions of up to about four times are achieved by regular transmission of short sections of digitally encoded speech—for example, sections of 10 ms duration separated by Intervals of 20 or 30 ms. An initial attraction of these techniques is that the interrupted speech is itself highly intelligible, although of very poor quality. At the receiver, the transmitted speech sections must be processed in order to synthesise a continuous speech signal of adequate quality. In Part One, investigations are described into simple synthesis techniques which operate by repeatedly outputting each transmitted section during the following interrupted period. From the effects of these techniques on speech quality, it is concluded that interruption is perceived aurally from the resulting temporal discontinuities in signal power, as well as from the associated spectral distortions. Synthesis techniques using adaptive waveform prediction are then considered, to achieve further improvements in the reproduced quality of Interrupted speech. These operate by creating an adaptive model of the speech signal from the received speech sections, and then using this to predict outputs during the interrupted periods. The predictor is described, and its performance illustrated by waveforms of processed speech and by informal judgements of its quality. Digital computer simulation techniques have been used widely in these investigations, and provide valuable experimental flexibility. Their use described in Part Two of this thesis. [PART TWO, Computer programs.] Some computer programs, and programming techniques for simulation of signal processing schemes are described. These were developed for use in the investigations described in the first part of this thesis.