The authors propose a new excitation model which can produce high quality speech at low bit rate (8 kbit/s) with low computational complexity. The excitation is represented by a sequence of pulses whose locations are coded by a small random codebook, and their corresponding amplitudes are solved such that the perceptual error between original and synthetic speech is minimized. They discover that the size of the codebook for the pulse locations can be made practically small without severe degradation. In order to further increase the quality of the speech, another source of excitation can be added. This excitation is obtained from a very small white Gaussian noise codebook so as to compensate the perceptual error between the original speech and the synthetic speech of the first excitation source. Since both codebooks are small in size, this scheme can be implemented with a low cost digital signal processor such as TMS320C25.
展开▼
