一种基于清浊音分类和参数冗余的语音差错隐藏算法  

A speech error concealment algorithm based on parameter redundancy and voiced/unvoiced classification

作　　者：肖勇[1,2] 张雄伟[2] 邹霞[2] 杨吉斌[2] 

机构地区：[1]61081部队,北京100000 [2]解放军理工大学通信工程学院,南京210007

出　　处：《信号处理》2009年第5期724-728,共5页Journal of Signal Processing

基　　金：江苏省“333高层次人才培养工程”资助项目;江苏省自然科学基金资助项目(BK2006001)。

摘　　要：为避免信道误码条件下低速率编码语音质量的严重恶化,提出了一种针对MELP语音编码的差错隐藏算法。在利用一阶马尔可夫模型描述帧间参数转移概率的基础上,算法构造了帧内增益与分带清浊音判决参数之间的转移概率;在语音合成时,提出了基于最大似然(ML)的清浊音分类模式保护;并对清音帧和浊音帧采用不同的参数重构方案。将算法与近年来针对MELP的两种差错隐藏算法进行了比较,测试结果表明在10^(-2)量级的随机误码的条件下,算法具有更好的差错隐藏性能。To avoid the severe degradation of low-bit rate coded speech when transmitted over noisy communication channels,an error concealment algorithm for MELP vocoder is proposed. One-step Markov chain is used to exploit the interframe and intraframe redundancy. The transfer probability between bandpass voicing parameter and the other parameters is presented. During speech synthesis, an algorithm based on maximum likelihood criterion is proposed to determine the voiced or unvoiced mode. Speech parameters are recon- structed according to minimum square error criterion during voiced frames, while reconstructed directly by mapping during unvoiced frames. Finally, our algorithm is tested under the scale of 10^-2 bit error rate. Compared with the two recently proposed error concealment algorithms, the performance of the new algorithm is more excellent.

关 键 词：语音编码 MELP 差错隐藏 清浊音分类 参数冗余 

分 类 号：TN912[电子电信—通信与信息系统]