基于HMM的移动音频编码丢帧隐藏方法

A packet loss concealment scheme based on HMM for mobile audio coding

针对恶劣移动音频传输环境下突发连续大量丢帧问题,本文提出一种基于HMM的丢帧隐藏方法,通过分析语音信号在更大范围的上下文关系的统计学变化来选择合适的丢帧隐藏策略。当包丢失时,基于HMM的恢复方法使用状态和密度函数信息,计算丢失帧参数的估计值。实验结果表明:提出的方法相比AVS-P10标准的语音编码器原有方法,客观语音测试PESQ平均分提高约0.33分,主观语音测试CMOS平均分能够提高约0.05分。

In order to deal with the problem of unexpected large number of continuous packet loss scenarios occur in poor quality of mobile voice transmission and audio communication service, a packet loss concealment scheme based on HMM for ACELP codec is proposed. The scheme utilizes HMM to decide the appropriate packet loss concealment strat- egy according to the statistical change of larger range of context in audio signal. When a packet loss occurs, the estima- tion is involved by recovery through status and probability density function. The experiment results show that the objec- tive listening test PESQ score achieves an increase of 0.33 points and subjective listening test CMOS score achieves an increase of 0.05 points when compared with the original scheme employed in AVS-P10 standard.