Joint optimization of adaptive predictive coding.
Seema A. Ranka · 1983
This work investigates the behavior of the residual signal in adaptive predictive coding (APC), which is one of the waveform encoding techniques. It also studies the variation of the predictor coefficients as a function of the structure of the predictor. Using continuous estimation of the predicted sample values of the speech signal, it is possible to force the residual signal to zero. Using the entropy of the continuously estimated prediction coefficients and of the residual, the structure of the adaptive algorithm is investigated. Results obtained using computer simulation show that the continuous estimation of predictor coefficients does result in a very low entropy for the residual distribution but predictor coefficients change dramatically from sample to sample, resulting in a very high entropy for the predictor coefficient distribution. Hence various different algorithms have been investigated to jointly optimize the APC. Prediction coefficients are changed only when the residual signal exceeds a particular value. Two different algorithms, which use this criterian to estimate the prediction coefficients are investigated. The results show that this represents a better approach towards reducing the number of bits/sample for the representation of the speech signal. The prediction coefficients do not change too much from sample to sample. It also keeps the residual distribution entropy low.