Backward adaptive lattice and transversal predictors in low-delay CELP coders

Abstract

This paper examines several backward lattice and transversal algorithms for updating the short-term prediction filter and the weighting filter in low-delay CELP (LD-CELP) systems operating at 16 kbit/s. We have considered block recursive adaptation structures and evaluated their relative merits both in terms of reconstructed speech quality and computational complexity. Comparisons in terms of performance are based on objective measures (signal-to-noise ratio, segmental signal-to-noise ratio and a frequency weighted signal-to-noise ratio) and on subjective results obtained from listening tests. Simulation results are presented and discussed for both ideal and noisy communication channels. Among the investigated adaptation schemes, the best performance in error-free channel was obtained with the exponential window lattice and the signal-driven lattice. However, a trade-off among robustness in noisy channels, performance under error-free conditions and computational complexity leads to the residual-driven lattice as an attractive algorithm to be used in LD-CELP structures.