P.1203 evaluation of real OTT video services

Abstract

Quality prediction performance of recently standardized parametric P.1203 models for real-streaming services: YouTube, Vimeo, Amazon Instant Video and a proprietary DASH-based streaming framework, is analyzed. In particular, a validation database comprising of bitstream traces from aforementioned services is used to evaluate the performance of P.1203 (mode 0 and mode 1) models. It is understood from the analysis that P.1203 models do capture the diverse encoding optimization strategies used in today's streaming applications to yield accurate prediction of long-term audio-visual quality, and the overall performance of P.1203 can be further enhanced using short-term Full Reference (FR) video-only models.