【作者】 王欣；

【导师】 别红霞；

【作者基本信息】 北京邮电大学 ， 信号与信息处理， 2014， 硕士

【摘要】 21世纪以来,随着网络技术的飞速发展,和多媒体业务的迅猛增长,人们试图利用网络承担多媒体业务的传输。网络电话(VoIP)就是基于此思想,通过网络实现语音通话。但是网络的非实时不可靠性使得通话质量无法保证。所以研究VoIP网络中能够有效的感知语音质量评估模型或方法,用以监控VoIP质量使之能保持一定的服务水准,是非常必要的。本文针对该问题,提出一种综合考虑传输损伤的客观评价模型,在得到总体得分的基础上,找出造成系统传输质量下降的关键因素,给予用户一定的指导作用。本文具体工作如下,其中(2)—(5)为主要创新部分。(1)研究当前话音质量评价方法,总结优缺点和测量范围；研究VoIP系统组成和协议结构,总结系统中影响话音传输质量的关键因素；(2)基于P.563和E-Model提出一种综合考虑传输损伤的评价模型,模型包含修正P.563模型和双向损伤两部分,该评价模型在多种损伤条件下均保持0.8左右的主观结果相关性；(3)根据E-Model中各损伤类损伤程度的计算方法,设计一套映射标准对不同损伤类的损伤程度统一量化为MOS分值,使得用户能够直观获得导致通话质量下降的关键因素；(4)提出一种基于信号相关性的收听者回声检测和参数估计方法,添加到P.563修正模型中,使修正后的P.563模型对于本文构建的收听者回声损伤语音库的评价相关性由原来的0.6101提高至0.875；(5)对一种基于频域相关的回声参数提取算法进行改进,在原有检测算法的基础上增加了双端通话状态判断、延时估计有效性判断和回声强度估计部分,并对执行速度进行了优化；研究并实现网络参数和延时参数的获取方法,并对提取效果进行验证；(6)构建标准测试语音库和损伤语音库,搭建网络仿真环境,并软件实现算法功能。更多还原

【Abstract】 Since the21st century, with the rapid growth of network technology and multimedia services, people try to use the network to transfer multimedia data services. VoIP is based on the idea that voice conversations over the network. But the unreliability of non-real-time network makes the speech quality can’t be guaranteed. Therefore, it is necessary that effective perceived speech quality assessment models or methods for monitoring VoIP quality to ensure the service level. For this problem, this paper proposes a objective measurement model considered transmission impairments comprehensively, which provides overall score, as well as identifies the key factors causing decline in the quality of the transmission system to give users some guidance. This article follows specific work, which (2)-(5) as part of the main innovations.(1) Study voice quality evaluation methods and summarize the advantages, disadvantages and range of current measurement. Research VoIP system components and protocol structure and summarize the key factors affecting the quality of voice transmission.(2) Combined P.563and E-Model, this paper presents a evaluation model considered transmission damage comprehensively. The model maintain. around0.8subjective correlation results under various conditions.(3) According to the degree of impairment type of E-Model, this paper designs a standard mapping of make a quantization of different kind of impairment to MOS, so that the user can get the key factors causing speech quality degradation visually.(4) This paper proposes a listener echo parameter detection and estimation method based on signal correlation and added it to P.563 model. The correlation of the modified P.563model to the listener echo speech degraded database constructed by this paper increased to0.875from the original0.6101.(5) Echo parameter extraction algorithm is improved, which add double-talk state judging, delay effectiveness judging and echo intensity estimation based on the original detection algorithm, and execution speed has been optimized. Research and realize the network and delay parameters acquisition method, and verify the effect of extraction.(6) Build standard test speech database and degraded speech database, build a network simulation environment, and develop related software.更多还原