VARIABLE RATE CODING OF SPEECH SIGNALS USING ADAPTIVE SUBBAND CODERS AND ADAPTIVE TRANSFORM CODERS

This dissertation is a report on research concerning Variable Rate Coding of Speech Signals using Adaptive Subband Coders (ASBC) and Adaptive Transform Coders (ATC). The main goal is efficient data compression or bit reduction of the coded speech samples. In this report we discuss the following three topics: (1) Variable Rate Coding Techniques. We present two different techniques: Fixed Signal-to-Noise Ratio (FSNR), and Fixed Noise (FN). In the FSNR technique, the bit rate (number of bits per unit time) varies such that the ratio of the signal energy to the noise energy is fixed. In the FN technique, the noise energy is fixed. It is shown that the FN coder performs better than the FSNR coder. This ressult is determined analytically as well as experimentally. (2) Overhead Processing for ASBC and ATC. Overhead information is the key to the coding scheme. We suggest the use of orthogonal matrices in overhead processing. The purpose is to reduce the amount of overhead information. This leads to bit reduction. These overhead processors can be used in fixed rate and variable rate coders. (3) Channel Noise versus Bit Error Probability. We examine the noise caused by a channel that is characterized by the bit error probability. It is shown that ASBC and ATC are not sensitive to the channel noise when the bit error probability of the channel is less than 10('-4). This result is true for both fixed rate coders and variable rate coders.