A 600BPS MELP vocoder with voice activity detection

Abstract

In the underwater communication, satellite communication, secure communication and other channels, the channel bandwidth is narrow and the channel condition is relatively poor. Therefore, higher quality and lower rate speech coding is needed. In order to improve the synthetic speech quality and save channel bandwidth, voice activity detection (VAD) technique is introduced to Mixed Excitation Linear Prediction (MELP) vocoder at 600bps in this paper. It can save channel bandwidth and reduce noise, coding rate and power consumption. In order to improve the accuracy of speech endpoint detection at low signal-to-noise ratio (SNR), noise reduction is adopted to improve SNR, and the VAD algorithm based on statistical model (STAT-VAD) is used. The MELP vocoder with VAD and noise reduction not only has good anti-noise ability and can improve robustness in random channel, but also can reduce the average coding rate and save channel bandwidth. The quality of synthetic speech can achieve the desired results at low SNR. In addition, the vocoder at 600bps with VAD can run in real-time on TMS320VC5510 DSP platform.