SELP语音编码模型的抗误码改进方案

Improvement of SELP Speech Coding Model for Anti-channel-error

受信道误码影响时,基于SELP模型的声码器合成语音出现大量刺耳的尖锐声,严重降低语音质量。研究发现,高能量的语音信号经分析滤波器产生余量信号,能量可能明显降低,量化值较小。信道误码可能使该参数在合成端出错明显增大,再经过合成滤波器后产生尖锐声。将能量参数从原始语音信号中提取,并改进合成算法中能量参数的用法。能量较高的语音信号量化值较大,发生误码时明显增大的可能性降低,能削弱尖锐刺耳声。

Synthesis speech of vocoder based on SELP（Sinusoidally Excited Linear Prediction） model would have some annoying sharp voice when channel error appears, which seriously reduces the speech quality. High energy speech would probably produce residual signal with very low energy from which the energy parameter would be extracted. As a result, the gain parameter of high energy speech may have low energy parameter. Channel error might make the gain parameter rise obviously for the decoder and annoying sharp voice would occur after the synthesis filter which could raise the energy of the input signal The energy parameter is extracted from the original speech and modify the usage of it for the decoder. Accordingly the gain parameter of higher energy speech would be larger which reduce the probability of the obviously rise of gain parameter when channel error occurs. Test result shows that the annoying sharp voice would be reduces obviously.