Design and implementation of a band saving analysis-by-synthesis scheme for speech applications

The main theme of this research is speech spectrum compression, basically, due to the increasing worldwide demand for lower transmission bit-rates for the speech communication systems. The primary problems associated with compression schemes are addressed: i) quality degradation, ii) severe information loss, iii) extensive time delay, and iv) complexity. The common approach used currently is based on efficient digital coding techniques, which might require a decision algorithm to detect the voicing property of the speech. However, this represents the major source of quality' degradation in present vocoders and much research is being devoted towards solving this problem. An alternative approach is followed in this thesis based on analog processing of the speech signal prior to transmission. Several schemes are proposed, evaluated, and compared both subjectively and objectively where every scheme is found suitable for a specific application and certain design assumptions are found to improve the performance of each system. The voicing property decision is avoided both in the analog processing stage and later in the digital coding stage whenever possible. The proposed systems have superior properties in terms of more control on the compression ratio over the present digital schemes. Moreover, the ability to combine the proposed schemes with low-rate coding techniques is also investigated and applied successfully. The principle of sub-band coding has been developed theoretically and applied practically in accordance with the dual sampling theorem. A new technique to multiplex and transmit the sub-band parameters is also introduced based on the spectrum scanning principle utilizing a time-invariant band-pass filter of variable center frequency. The value of the center frequency varies in discrete steps following a discrete time sweep voltage. The latter is designed to make use of the slow-varying property of the speech signal. The aforementioned schemes have been tested experimentally under different compression ratios and transmission bit-rates by implementing a hardware system and/or a simulation system for each proposed scheme. Simulation programs for most of the
schemes are also developed and compared with the practical systems for the purpose of objective system evaluation. These programs have been implemented under the support of Matlab package version 5. Typical test results show that the speech bandwidth can be compressed to about 50% without an annoying degradation in quality. Additional compression and redundancy removal is also achieved in the coding stage to transmit a speech signal at a rate of about 5.2 kbit/s with acceptable quality.