Speech coding based upon vector quantization

A procedure for the vector quantization of speech is developed and demonstrated to have a dramatic advantage over scalar quantization when compared on an rms log spectral measure basis. Informal listening has verified that equivalent rms log spectral measures are also, roughly speaking, perceptually equivalent. That is, at a specified dB level such as 1.8 dB, no marked preference occurs between a 35 bits/frame scalar-quantized speech sample and a 10-bit full search vector-quantized sample.