Packet-layer Quality Assessment Method For Networked AAC Audio Streams

To satisfy the requirement of high service quality of the networked audio andprovide advanced experience for consumers, it is necessary to monitor the audio qualityin real-time in the application of networked audios. Packet-loss over the network has aserious effect on the audio quality, and therefore it is important to evaluate the distortionof the transmitted audio stream caused by packet loss. According to the characteristicsof networked audio, this thesis focuses on packet-layer audio quality assessment.Through a large number of experiments, the impact of encoded frames on theoverall audio sequence is evaluated. Based on the observations, a temporal poolingmodel is proposed to predict the audio quality for applications over two differentnetworks, i.e., the Internet and the Cable Digital Television (CDT). The proposed modelis able to evaluate the subjective experience of consumers on various types ofpacket-loss. Considering the impact of packet-loss on a encoded frame, the distortionsdirectly caused by packet-loss and due to error propagation are analyzed respectively,based on which the quality of various frames are predicted. Then the quality of asegment of the sequence is obtained using the proposed temporal pooling model. Finally,for quantifying the impact of packet-loss on the overall audio quality, this paper predictsthe audio quality through2-level temporal pooling. Experimental results show that theproposed temporal pooling model is effective.The proposed temporal pooling model is then applied to evaluate the quality ofnetworked audio stream over CDT. By analyzing the structure and function of TSpackets widely used in CDT, a novel frame quality assessment method suitable for theTS packet stream is proposed, based on which the audio quality can be obtained by2-level temporal pooling model. The experimental results demonstrate that the adaptedtemporal pooling model has an advanced performance.