引入内容特性分析的包层语音质量评价模型  

A Packet-layer Model for Speech Quality Assessment Introducing the Analysis of Content Feature

作　　者：江亮亮[1] 李雪敏[1] 杨付正[1] 杨旭[1] 

机构地区：[1]西安电子科技大学综合业务网理论及关键技术国家重点实验室,陕西西安710071

出　　处：《四川大学学报（工程科学版）》2013年第3期103-107,共5页Journal of Sichuan University (Engineering Science Edition)

基　　金：国家自然科学基金资助项目(60902081);中央高校基本科研业务费专项资金资助项目(72115612);高等学校学科创新引智计划资助项目(B08038)

摘　　要：为了实现对网络语音质量的实时监控,提出一种包层语音质量评价模型。该模型无需介入数据包的载荷部分,只利用数据包的头信息评价语音质量。首先通过分析包头信息区分出语音段和静音段,获取语音段的编码参数和丢包参数,然后根据语音段的编码参数预测编码失真,在此基础上利用语音段的丢包参数评价丢包引起的失真,从而得到语音流的总质量。实验结果表明,相比于国际标准G.107中的E-model,提出的模型得到的语音质量评分与PESQ算法评分的皮尔森相关系数平均提高0.041 2,均方根误差平均降低0.045 1。A packetlayer model for speech quality assessment was proposed to monitor the quality of networked speech. Without resor ting to any mediarelated payload information, the proposed model predicted the speech quality only in terms of the information provided by packet headers. First, the analysis of content feature was performed to locate the voiced segments and silence segments by analyzing the information from packet headers, and the coding and packet loss parameters of voiced segments were obtained. Then the coding dis tortion was estimated according to the coding parameters of voiced segments, based on which the overall speech quality was further eval uated by taking account of the impact of packet loss. Experimental results showed that the proposed model could get an increment about 0. 041 2 in Pearson correlation coefficient （PCC） and a decrement about 0. 045 1 in root mean squared error （RMSE） compared with the Emodel proposed in ITUT recommendation G. 107.

关 键 词：语音编码 语音传输 服务质量 丢包 

分 类 号：TN912.3[电子电信—通信与信息系统]