一种基于信道软译码的差错隐藏算法  

Speech error concealment algorithm based on soft-decision channel decoding

作　　者：肖勇[1,2] 张雄伟[1] 邹霞[1] 谢贵武[1] 

机构地区：[1]解放军理工大学通信工程学院 [2]解放军61081部队

出　　处：《解放军理工大学学报（自然科学版）》2009年第2期113-117,共5页Journal of PLA University of Science and Technology(Natural Science Edition)

基　　金：江苏省自然科学基金资助项目(BK2006001)

摘　　要：为满足恶劣无线环境下低速率语音通信需要,针对混合激励线性预测(MELP)提出了一种基于信道软译码的差错隐藏算法。利用软译码所得对数似然概率比估计比特错误概率,提出了利用比特错误概率和信源残留冗余,基于最大后验概率(MAP)估计合成端语音清浊音模式。对不同的清浊音模式,选择不同的参数重构方案,浊音帧时,利用对数似然概率比基于最小均方误差(MMSE)准则进行参数重构;清音帧时,硬判决对数似然概率比,由硬判决结果通过映射直接重构参数,在信道编码为递归系统卷积码和并行级联卷积码情况下,采用PESQ测试了提出的差错隐藏算法误码条件下合成语音质量。测试结果表明,与近年来提出的针对MELP的差错隐藏算法相比,该算法具有更好的性能。To meet the demands of low-bit rate compressed speech in wireless channels, a speech error concealment algorithm based on soft-decision channel decoding for MELP（mixed-excitation linear predictive） was proposed. The bit error rate was estimated from the log-likelihood ratio of soft-decision decoding. The voiced/unvoiced mode was determined according to a maximum posterior probability criterion, exploiting the soft information in log-likelihood ratio and the residual redundancy in compressed speech. Parameters were reconstructed according to the voiced/unvoiced mode. During voiced frames, parameters were reconstructed according to MMSE （minimum square error） criterion, while reconstructed directly by mapping during unvoiced frames. Finally, the performance of this algorithm was tested through PESQ considering the channel code as recursive systematic convolutional code and parallel concatenated convolutional code. Compared with the two recently error concealment algorithms for MELP, the new algorithm has better performancs.

关 键 词：信道软译码 差错隐藏 清浊音模式 最大后验概率 

分 类 号：TN912.3[电子电信—通信与信息系统]