A variable-bit-rate speech coding algorithm based on enhanced mixed excitation linear prediction

In order to improve the channel bandwidth utilization of voice communication, a variable bit rate speech coding algorithm based on enhanced mixed excitation linear prediction (MELPe) is proposed in the paper. In voice communication, only about 40% of the time is occupied by talking, whereas the rest is engaged by silence or background noise. In addition, unvoiced frame usually requires less transmission rate than the voiced one in low bit rate speech coding algorithms. Therefore, always using the same coding bit rate for speech coding is a waste of channel resource. In this paper, the input signal is divided into speech and silence by using voice activity detection (VAD) technology. And the speech frames are divided into voiced frame or unvoiced frame. They use different coding rates for speech coding and data transmission. All of the parameters are encoded, transmitted and decoded in voiced frame. Only gain parameters, LSF parameters, pitch parameters and overall voicing are encoded, transmitted and decoded in the unvoiced frame. Furthermore, only the gain parameters and the first level LSF parameters are encoded, transmitted and decoded in the silence frame. When about 40% of the time is occupied by talking, compare with the traditional 2.4 kbps MELPe vocoder, the average coding rate of the proposed variable bit rate vocoder can reach 1.33 kbps. But they can achieve the same quality of synthetic speech. Experimental results show that the proposed method reduces the average coding rate, and the synthetic background noise has good comfort on the subjective sense of hearing.