Statistical Measurement Approach for On-line Audio Quality Assessment

The quality of audio in IP telephony is significantly influenced by the impact of packet loss rate, burstiness and distribution on audio compression techniques. On-line audio quality assessment is important to provide real-time feedback to end-to-end Internet audio transport protocols to increase the reliability and quality of the audio session. In this paper we propose a novel passive statistical measurement framework, Audio Genome, that can deduce the audio quality of an on-going Internet audio for many different codecs under any network loss condition at real-time. We use multiple polynomial regression as the modelling technique to accurately characterize the audio quality for codecs under any loss scenario. Our approach is easy to deploy and guarantees high computational speed. Our extensive evaluation experiments, that include large simulation scenarios, show that our framework is accurate and viable for adaptive real-time audio mechanisms.