A channel-optimized predictive speech coder

A method for channel-optimized vector quantization of the predictor coefficients and the residual signal of a predictive speech coder is presented. Using the residual energy distortion measure for the long-term and short-term predictors and the spectrally weighted squared error distortion measure for the residual signal, an algorithm for channel-optimized vector quantizer and channel detector design is derived. The technique is applied to a fully vector-quantified long-term predictive ADPCM coder with in-loop short-term prediction and microphone and speaker adaptation filtering. Simulation results for 4PSK and 8PSK modulation on the additive white Gaussian noise channel are presented. The results indicate that by using the proposed techniques, a large improvement in the reproduced speech quality can be achieved for noisy channels while maintaining nearly the same noiseless channel quality