Sub-band Unvoiced/Voiced parameter extraction and efficient quantization for speech signal

In Mixed Excitation Linear Prediction algorithm (MELP), the sub-band Unvoiced/Voiced parameters play an important role in improving the naturalness of synthetic speech. However, the coding efficiency with five bits per frame brings difficulties for very low bit rate speech coding. In this paper, the three consecutive MELP frames are grouped into a super-frame, and the fifteen dim sub-band Unvoiced/Voiced parameters are quantized. Through counting the Unvoiced/Voiced distribution probability and optimizing the codebook designed by the distortion measure, it is implemented that every fifteen dim Unvoiced/Voiced vector is quantized efficiently with three bits for each super-frame. Simulation results show that the intelligibility and naturalness are efficiently maintained for synthesis speech, and the quantization scheme can be widely applied to speech coding algorithm below 600bps.