基于SELP声码器的连续丢包隐藏算法  

Consecutive packet loss concealment algorithm based on SELP vocoder

作　　者：计哲[1] 徐敬德[1] 崔慧娟[1] 唐昆[1] 

机构地区：[1]清华大学电子工程系,微波与数字通信技术国家重点实验室,北京100084

出　　处：《清华大学学报（自然科学版）》2010年第12期2003-2006,共4页Journal of Tsinghua University(Science and Technology)

基　　金：国家自然科学基金资助项目(60572081)

摘　　要：针对在连续丢包情况下声码器合成语音质量较差的问题,提出了一种特征参数的分模式线性预测技术。该方法利用参数的短时相关性,以子带清浊音参数为模式信息,计算各特征参数在不同模式下的预测系数,并根据获得的分模式预测系数用上一个正确接收帧的特征参数预测当前丢失帧的参数,最后用恢复的参数重建丢失语音帧。测试结果表明：当丢包长度的范围在75-200 ms时,与传统的抗丢包处理算法相比,该方法能够将合成语音的平均意见得分（mean opinion score,MOS）提高0.03左右。Vocoder quality deteriorates greatly when consecutive packet loss occurs.This paper presents a mode based linear algorithm to predict the characteristic parameters during packet loss.The algorithm utilizes a short-time correlation of the parameters,with the sub-band voiced/unvoiced parameters as the mode information to calculate the time-domain correlation coefficients of each parameter for different modes.The algorithm predicts the characteristic parameters of the lost frame using the parameters of the preceding correctly received frame to reconstruct the lost frame.Simulations show that when the lost packet length ranges between 75 and 200 ms,the algorithm increases the objective mean opinion score（MOS） of the synthesized speech by 0.03 compared to the traditional packet loss concealment method.

关 键 词：语音编码 丢包隐藏 特征参数 线性预测 

分 类 号：TN912.32[电子电信—通信与信息系统]