Vector-adaptive vector quantization with application to speech coding

A vector-adaptive vector quantization (VAVQ) scheme, which may be viewed as a generalization of gain-adaptive vector quantization, is described. The proposed scheme adjusts each component of the encoding signal vector according to a statistical estimate of the signal characteristics. The VAVQ scheme can cope with large input dynamic ranges. It can be used in either the time domain or the transform domain, and exhibits approximately 4 dB improvement in segmental signal-to-noise ratio (SNR) over the fixed VQ.>