Audiovisual Quality Estimation for Video Calls in Wireless Applications

Mobile/wireless multimedia applications (e.g., video calls and IPTV) have gained great momentum in recent years. An important issue is to monitor/predict overall audiovisual quality, instead of audio-only or video-only quality, non-intrusively for technical or commercial reasons. Previous audiovisual modeling research mainly considered application parameters (e.g., codec and send bit rate). Little attention has been paid to how network parameters, e.g., Packet Error Rate (PER) affect audiovisual quality. The aim of this paper is to explore methods to predict audiovisual quality objectively for video calls in wireless applications. The contributions of the paper are twofold. Firstly, we present subjective test results on how audio and video contribute to overall audiovisual quality and develop models to reflect this relationship. Secondly, we investigated how network parameters (e.g., PER) and application parameters, e.g., video Frame Rate (FR) affect overall audiovisual quality. We developed a regression model to predict audiovisual quality from PER and FR which can be used to monitor/predict audiovisual quality non-intrusively. We also explore the possibility to predict audiovisual quality from full-reference voice and video quality metrics (i.e., PESQ and PSNR) and from combined PESQ/PSNR and network/application parameters. The different predication accuracy obtained from these models (accuracy from 84% to 93%) indicates the complex attributes in audiovisual quality prediction. An extended Evalvid/NS-2 platform is developed to support simulation of video calls over wireless networks.