PROCESS FOR VECTOR QUANTIZATION, ESPECIALLY OF VOICE SIGNALS

摘要

Vector quantization of voice samples uses two codebooks (CB1, CB2) into which are entered the fixed excitation vectors and pitch parameters of a prediction filter, which are updated at time intervals. To improve voice quality, in each case the two best - defined in terms of an error criterion - vectors from the adaptive codebook (CB2) are combined with all vectors of the fixed codebook. From the aggregate of combined vectors is selected that value which best corresponds to the original voice sample.