Incentives Based On Silence Detection Technology With Voice Vocoder

Multiband Excitation (MBE) codec is a better choice among the speech coding algorithms at low rate. It can synthesis better speech than the traditional codec at rate of 2.4～4.8Kb/s. Its synthetic speech sounds naturally and can endure more noise. The paper took MBE codec as the mayor object of research. Detailed research and experiments had been done from the two aspects of speech analysis and synthesis. In the meantime, several classical algorithms of voice activation detection(VAD) had been researched and simulated carefully in this paper. The simulations proved that: The VAD algorithm based on entropy worked well than others and can detect the speech/mute frames relative accurately.Considering the VAD effective detection of voice and mute, the VAD algorithm was applied to the MBE codec, in order to reduce the code rate. Before codering, the VAD algorithm was implemented firstly to distinguish the speech/mute frames, then different speech coding algorithms were applied to them. MBE codec was executed during the speech frames. Only the mean of amplitudes was sent to the decoder during the mute frames. Similarly, different speech decoding algorithms were implemented to the speech/mute frames .The MBE codec based on VAD algorithm was simulated by MATLAB and C. The concrete implementation and results of the improved MBE algorithm was shown. The simulations showed that: The MBE algorithm based on VAD effectively decreased the rate and computational time, and the synthetical speech was very similar to the original on the time region. The synthetic speech was clear and understandable. At the same time, this paper improved the FFT algorithm of the MBE algorithm and proposed an improved FFT pruning algorithm. The improved FFT pruning algorithm was run in the C5402 Device Simulator of CCS. The results of exam show that: The computational consumption of it was obviously less than otherFFT pruning algorithms.