A hybrid wavelet-binary pulse excitation approach for high quality CELP speech coding

A hybrid wavelet-binary pulse excitation (WBPE) model for CELP speech coding is reported. The WBPE arises as the result of some preliminary experiments, for modelling the residual predictive error using a wavelet modelling process. In a general WBPE model, the excitation is modelled as a combination of the long-term prediction signal and wavelet functions spaced by the optimum closed-loop pitch distance and located using the pitch adaptive codebook approach. Finally, a constrained stochastic excitation is added using a binary codebook. The initial results of an example of the WBPE targeted at 6.3 kb/s are also compared with the conventional CELP coder at 8 kb/s using Gaussian, sparse and ternary codebook population