Non-intrusive single-ended speech quality assessment in VoIP

Abstract

Evaluating speech quality in voice over Internet protocol (VoIP) in a non-intrusive manner is challenging, because it relies on a degraded speech signal only. In this paper, a parametric, non-intrusive VoIP speech quality assessment algorithm is proposed, which adopts a three-step strategy, impairment detection, individual effect modeling and an overall model. Mainly based on voice payload analysis, the algorithm also combines Internet protocol analysis approach and the ITU-T E-model. It quantifies the individual contributions to speech quality from several major VoIP impairments, including packet loss, temporal clipping and noise. Also, an overall assessment model is developed. The performance is evaluated through intensive simulations, and the results show that the algorithm is effective and accurate. For the overall model, the correlation between prediction and measurement is 0.90; the root mean square error (RMSE) is 0.27 mean opinion score (MOS). The algorithm aims to be implemented at the receive-end media gateway or IP terminal, for identifying the root causes of speech quality degradation as well as quality assessment in VoIP.