Quality assessment for networked speech based on active packet detection

In this paper the assessment of perceived speech quality under packet loss is investigated. Since packet loss would impair the perceived speech quality differently when lost packets are located in different segments in the speech signal, e.g. in speech or silence segments, a new detection method is first proposed to accurately determine the active packets which are related to speech signals only. Based on the detection of active packets, the packet loss rate within speech segments can be determined, which is further incorporated in the E-model for speech quality assessment. The efficiency and accuracy of the proposed detection method and the method for speech quality assessment have been verified by experimental results. Performance evaluation shows that the proposed method based on the active packet detection outperforms the original the E-model by over 8% in accuracy of prediction of the subjective speech quality.