用户体验质量视角下的端到端语音质量异常检测系统

End-to-end evaluation system for detecting special voice deteriorations from view of user experience

为在商用网络中从用户体验角度检测单通、串话、回声和杂音等语音异常,在重构P.862和P.563算法基础上,实现一种语音异常检测算法。通过分析P.862和P.563相关参数的物理意义,确定对语音异常较敏感的参数;分析各参数和语音异常之间的逻辑关系,进一步减少每次的参数计算量,降低算法复杂度。基于商用网络的实验结果表明,由于较低的复杂度,该算法可用于移动网络的实时商用测试,保持了较高的检测精度。

To detect the special speech deteriorations such as single talk,cross talk,echo and voice noise from the view of user experience,a detection algorithm was designed on the basis of P.862 and P.563.Through analyzing the physical interpretation of the parameters of these two algorithms,the parameters sensitive to speech deteriorations was identified.The amount of computations was further reduced via building the logical relationship among these parameters.As a result,the complexity of the proposed algorithm was reduced significantly.In the commercial networks,the experimentation results demonstrate that the proposed algorithm is competent for real-time evaluation for the mobile networks.