Generalized predictive trellis coded quantization of speech

Trellis-coded quantization (TCQ) is incorporated into a generalized predicted (noise feedback) structure for encoding sampled speech. Adaptive residual encoding and adaptive prediction are used to obtain signal-to-noise ratios (SNR) in the range of 17.5 to 21.5 dB for encoding sampled speech with 2 bits/sample (16 kb/s). Spectral noise shaping is achieved with this coding structure by using a bandwidth-expanded version of the predictor as a noise feedback filter. This has the effect of reducing the SNR of the encoding by a small amount but improving the perceptual quality of the reconstructed speech. The result is of excellent communications quality. A modified Viterbi algorithm is used for the trellis search to allow flexibility in the choice of symbol release rule. The effects of varying the encoding delay and the number of symbols released per trace-back on system performance and complexity are investigated. It is shown that excellent SNR performance can be obtained with modest encoding delays.<>