Analysis and evaluation of an adaptive silence deletion algorithm for compression of telephone speech

This thesis is concerned with the analysis and evaluation of adaptive silence deletion as a means to compress telephone voice signals bandlimited to the range 200-3400 Hz. Speech is accompanied by noise arising from various environmental factors such as poor reception, interference of radio signals from mobile or cordless units, audible mechanical or social activities in the surroundings, and the conventional crosstalk and hum in the telephone system. A speech compression system based on significant modifications to an existing silence deletion algorithm has been implemented. Effects of the various system parameters on the operation of the system, as applied to telephone speech samples, are studied and analyzed graphically. Quality of the speech compression is assessed with subjective listening tests. With minimal algorithmic complexity and delay, the application of silence coding together with 4-bit ADPCM speech coding can compress uncoded telephone speech from an original bit rate of 128 kbps down to 16 kbps. Analysis of system performance shows that a processing frame size of 8 to 16 milliseconds yields the best combination of speech quality and compression efficiency. A set of system parameters is found to give robust performance in a wide range of operating environments, with different or varying speech and noise levels. Good playback quality resulting from compressed speech recorded in quiet and also in noisy environments is achieved at 50 percent compression, equivalent to half the bit rate of ADPCM.Applied Science, Faculty ofElectrical and Computer Engineering, Department ofGraduat