A Low Bit Rate Speech Codec Using Mixed Excitation Linear Prediction for Private Mobile Radio

摘要

The authors describe a low bit rate speech codec developed for the next generation of narrow-band (3 kHz/voice channel) private mobile radio. The speech encoding speed under the proposed speech codec is 1.6 kbit/s, and the speed after communications path encoding is 3.2 kbit/s. The authors propose a method using the speech spectrum envelope information to determine the mixing ratio for pulses and noise in each frequency band used to generate the mixed excitation signal in the Mixed Excitation Linear Prediction (MELP) vocodor model used as a speech encoding method. For the communications path encoding method, the authors assume a burst error of 7% for the worst communications environments in private mobile radio. In order to improve resistance to transmission path errors, the authors create soft decision demodulation data while varying the quantization step width until the errors are no longer detected or a designated number of repetitions is reached for the frame determined as having errors based on the results of error detection during communications path decoding. They then propose a method to repeat soft decision Viterbi decoding. The authors perform subjective evaluation experiments, and confirm the effectiveness of the proposed speech codec in a private mobile communications environment.