Hybrid vector quantizers for low bit rate speech coding applications

Abstract

Vector quantization techniques play a dominant role in compression of speech signals. There exists a variety of vector quantization techniques, each technique has its own advantages and disadvantages and no vector quantization technique is perfect in all aspects till now. This thesis deals with enhancing the performance of the existing vector quantization techniques using hybrid methods. In this thesis two hybrid vector quantization techniques are proposed which are developed from the existing vector quantization techniques. To find the performance of these vector quantization techniques, they are used in linear predictive coder to reduce the bit-rate of the speech signal, without any considerable loss in the quality of reconstructed speech signal. The performance of the vector quantizer is evaluated in terms of the spectral distortion measured in decibels (dB), computational complexity in flops per frame and memory requirements in floats.