Adaptive prediction with quantized data

One of the most important existing problems involving adaptive prediction with quantized data is the design of an adaptive predictor for a differential pulse code modulation system. Four different algorithms are compared: a least squares lattice algorithm, a least mean square lattice algorithm, a transversal structure Kalman algorithm, and a least mean square transversal algorithm. The data base for the comparisons are five sentences spoken by male and female speakers. Based on objective measures and subjective listening tests the least squares lattice algorithm yields the best performance. For noiseless channels, all adaptive algorithms outperform a fixed predictor. When the channel is noisy. the lattice algorithms are clearly preferred over the transversal forms.