截幅失真对低速语音编码的影响分析及改进  

Study on the Influence of Clipping on Low Bit Speech Coding

作　　者：吴彭龙 邹霞 孙蒙 张星昱 Wu Penglong;Zou Xia;Sun Meng;Zhang Xingyu(Army Engineering University of PLA, Nanjing, Jiangsu 210000,China)

机构地区：[1]陆军工程大学指挥控制工程学院,江苏南京210000

出　　处：《信号处理》2020年第3期426-438,共13页Journal of Signal Processing

基　　金：江苏省优秀青年基金(BK20180080)。

摘　　要：截幅失真会影响语音编码质量,特别在低速率语音编码条件下,截幅语音不再符合人发声模型,编码语音质量严重下降。为了研究截幅失真对低速率语音编码的影响,从截幅语音编码参数提取和截幅语音编码质量两个方面进行了分析。采用偏离度衡量低速率语音编码参数提取的准确性,编码参数包括LPC、基音周期和清浊音。在不同截幅程度下,分析了各种参数错误分布、错误类型和错误原因。采用客观感知语音质量评估PESQ打分评估截幅语音编码质量。针对常用截幅修复方法在截幅程度较大时修复性能下降严重的现象,提出采用两种改进型截幅修复方法对截幅语音进行修复。实验结果表明,改进的截幅修复方法能有效提高截幅程度较大时的低速语音编码质量。Clipping can affect the quality of speech coding,especially under the condition of low bit speech coding.Clipped speech no longer conforms to human voice model,and the quality of speech coding for clipped speech is seriously degraded.In order to study the effect of clipping on low bit speech coding,two aspects of the parameters extraction of clipped speech and the quality of clipped speech coding are analyzed.Deviation degree is adopted to measure the accuracy of parameter extraction,and the parameters including LPC,pitch and voicing decision.The error distribution,error types and error causes of various parameters are analyzed under different clipping degrees.The coding quality of clipped speech are evaluated by objective perceptive speech quality assessment(PESQ).In order to solve the problem that the performance of the traditional declipping methods is decreased when the clipping degree is large,it is proposed to use two improved declipping methods to repair the clipping speech.The experimental results show that the improved declipping method can effectively improve the performance of low speed speech coding when the clipping degree is large.

关 键 词：语音编码 截幅修复 混合激励线性预测编码器 

分 类 号：TN912.3[电子电信—通信与信息系统]