Low-rate Speech Coding Algorithm Based On Wavelet Transform

In the past 50 years, speech encoding technology has been made quite great progress. Up to now, the speech quality of encoding algorithm above 4 kbps is almost equal to that of PCM encoding algorithm. And it is being brought into the practical period. Currently, low bit-rate speech encoding algorithms are focused on high quality and low time delay of the vocoder below 4 kbps.MELP speech encoding algorithm is based on the traditional LPC model, which replaces duality excitation with mixed excitation and gets better result on the bit-rate of 2.4 kbps. To get better quality of synthesized speech, the algorithm including speech analysis, parametric encoding, and speech synthesis is researched completely, which is based on the core of the MELP algorithm. A new method of frequency subband separation using wavelet transformation is proposed in the paper, which is applied in the extraction of pitch parameter, bandpass voicing parameter, and residual signal. To reduce the encoding bit-rate, a vocoder's scheme with the bit-rate of 1.0 kbps is proposed, which uses the methods of both superframe model and joint vector quantization.The two algorithms mentioned above are simulated on the Matlab ver.6.5. The comparison with original speech waveform and the result of real audition test show that the synthesized speech is satisfing. Finally, the above two algorithms' implementation using DSP is studied.