Low delay audio coder using adaptive vector quantization

Nowadays, there is not a low delay audio coding standard that achieves nearly-transparent quality. The ISO-MPEG standards are profusely used in audio for high quality coding. They use perceptual models that require high frequency resolution. In applications where the delay is a critical parameter, i.e. when the use of a feedback channel is important, the ISO-MPEG algorithms are not suitable. In this paper we present a new coder that responds to a subband ADPCM structure and incorporates adaptive vector quantization, requiring only 2 bits/sample. Each vector component (prediction error) is weighted according to its estimated standard deviation. The introduced coding delay is less than 2 milliseconds, and is only due to the perfect reconstruction filter bank. In order to obtain the best configuration, results have been obtained varying the number of bands, the number of bits in the quantizer and the analysis filters.