FORWARD ADAPTIVE SPEECH CODING WITH LOW BIT RATES AND VARIABLE WORD LENGTH

In this paper, two scalar quantizers for the memoryless Laplacian source with low number of levels are designed and discussed. The nonuniform quantizer is designed according to the Lloyd-Max’s algorithm since it can provide an optimal performance in the minimum distortion sense. Two variants of the uniform dead-zone quantizer are designed according to the criterion of minimal distortion and the simultaneous criterion of minimal distortion and minimal bit rate. Joint design of quantizer and Huffman encoder is considered in all proposed solutions. In addition, forward adaptation of the observed quantizers is performed on frame-by-frame basis. The best performance from the point of practical implementation is obtained using a uniform dead-zone quantizer that satisfies the criterion of minimal distortion and minimal bit rate at the same time. Moreover, the theoretical results are verified via the experimental results obtained on a real speech signal