语音编码抗信道误码保护算法

Channel error resilience protection algorithm for speech codec

为了提高在高误码率窄带无线信道下的合成语音质量,提出了一种信源信道联合编解码保护的语音编码抗误码算法。该算法在编码端利用编码后的冗余度进行 BCH编码和奇偶校验以保护对语音合成质量影响较大的参数;在解码端对清浊音参数采用分支判决和改进的最大后验概率算法进行恢复,在浊音帧对线谱对(LSP)参数进行基于信源信道联合特性的线谱对参数差错后处理,在清音帧采用 BCH 解码和前向替代。该算法在不消耗任何额外带宽且无算法延时的条件下可以显著提高语音编码抗信道误码能力和恶劣信道条件下的合成语音质量。仿真实验显示,在较高信道误码率下平均谱失真降低了25%～36.1%,平均意见得分(MOS)提高了12.33%。

The paper proposes a novel error resilience protection algorithm for speech codec to enhance the reconstructed voice quality under high error rate narrowband wireless channels. At the encoder side, the proposed algorithm utilizes the residual redundancy inherent in the coded voice for performing BCH and parity check to protect the parameters which are iraportant to the reconstructed speech quality. At the decoder side, the unvoiced and voiced parameters of speech codec are and then the line spectrum pair （LSP） parameters are recovered in the voiced frame using joint source and channel characteristics. BCH decoding and error replacement are utilized in the unvoiced frame. The proposed algorithm requires no extra bandwidth and algorithm delay while greatly enhancing the error resilience ability of speech codec and the reconstructed voice quality under severe channel conditions. The simulations showed that the MOS （mean opinion score） was improved by 12.33% and the spectrum distortion was reduced by 25% -36.1%.