甚低速率语音编码中的高效模拟退火算法研究

为了解决600bps甚低速率语音编码中码本矢量量化信噪比低以及抗信道误码能力差的难题,本文提出了一种改进的模拟退火算法,在600bps甚低速率语音编码中得到了成功的应用。首先,采用改进的模拟退火算法设计语音编码参数矢量量化码本。然后,采用模拟退火算法分配码本序号,以此提高在信道误码条件下语音编码算法的鲁棒性。仿真结果表明本算法的量化信噪比与经典的LBG算法相比改善了0.4dB左右,平均误码失真与采用随机分配码本序号的方法相比改善60%以上。非正式主观试听表明600bps语音编码算法的合成语音具有良好的可懂度和自然度,并且在1%信道误码条件下仍然具有良好的性能。

Research of an Efficient Simulated Annealing Algorithm in the Very Low Bit Rate Speech Coding

An improved simulated annealing algorithm is proposed and applied to the 600bps speech coding to solve the problem of low quantization SNR and the poor performance in noisy channels. First, the algorithm is used to design the codebook of the speech coding parameters. And then, simulated annealing algorithm is employed to assign binary codewords to the codebook to improve the speech quality in noisy channels. Simulation results show that the quantization SNR of the algorithm is about 0.4dB better than that of the traditional LBG algorithm, and the improvement of average distortion per bit error exceeds 60% compared to randomly selected codewords. The result of informally subjective listening test indicates that the intelligibility and naturalness of the 600bps vocoder is quite well. Furthermore, the speech coding algorithm is robust in a noisy channel with 1% BER.