QW2002 Paper 7T1 Dr. Chris Overton (Keynote Systems) & Dr. Simon Robins (Genimedia) Streaming Media Quality: Orphan Child of "Old Media" and Internet Flakiness

Key Points

• Streaming media quality IS rigorously measurable in MOS-type metrics
• You also need low-level operational metrics tied to discrete events
• The challenge lies in bridging the gap - attemptable from both directions

Presentation Abstract

For standardized old media such as television, telephone, or even radio, there are mature infrastructures for quantifying quality, based on correlation to Mean Opinion Scores (MOS), which average over panels of human judges. This has not yet become widespread for streaming media, due to its variety, rapid evolution, novel degradations, and market immaturity. Nevertheless, we present rigorous pixel-based metrics that correlate well with MOS (Genimedia approach), as well as metrics built upward from low-level operational events (Keynote.) We discuss several important tradeoffs in the two approaches (interested audience, operational and SLA usability, intrinsic meaning, maintainability), as well as how we see the two approaches being brought together in the future. Many metrics are demoed under actual video degradations.

About the Author

Chris Overton is Keynote Statistician & Quantitative Architect. Over the last decade, he has consulted as a statistician in industry and in academic biomedicine, as a software architect and developer, and in business model development. He founded Crazy Tulip Corp. to build knowledge modeling software systems.
His responsibilities at Keynote include algorithm and tool design, data analysis & interpretation, internal & external education, and serving as academic liaison. He architected Keynote’s SLA reporting engine and has helped several large companies build SLAs, including for streaming media both on the provider side and on the customer side.
Chris is the principal architect of Keynote’s streaming media metrics and has published and lectured on related topics. He got his pure math PhD from Stanford in ?96 and has taught there and at the University of San Francisco.

Simon Robins has been working in software development for over 10 years, and joined Genimedia as Principal Engineer on its foundation at the start of 2001. Genimedia is dedicated to perceptual quality measurement for the monitoring, management and control of the production and delivery of digital media. He now leads the technical marketing team at Genimedia. He has a degree in theoretical physics from the University of Liverpool and a PhD in particle physics from London University.