OpenTok MOS Estimator
Background
Each OpenTok client SDK includes network stats callbacks on the subscriber to allow programmatic access to a subset of the WebRTC getStats dictionary. While the stats seem interesting on their own, this project aims to digest available getStats data into a single, simple to use metric, a numerical 1-5 estimate of the mean opinion score (MOS) of a given subscriber. This data differs from other existing tools by being made available on-demand, in real-time, and on the endpoint, rather than through a server-side data process.
Usage
Specific usage details for each platform SDK are available in respective subproject directories. Overall, the workflow is the same for each -- a new module is introduced to either subtype or attach to a normal OpenTok Subscriber.
Notes
-
Scoring algorithm is loosely based on the ITU-T E-model.
-
Audio scores are calculated as primarily a function of packet loss, and RTT is gathered when possible, depending on platform availability.
-
Video scores are calculated primarily as a function of bitrate, where target bitrates are considered based on the assumed resolution of received video. Low resolution video does not currently incur a score penalty, although there are compelling reasons for doing so.
-
Both scores are calculated periodically (configurable), and a requested score is given as the mean of all calculated scores. An overall score for the subscriber is the minimum of the two audio/video scores, or whatever is available for the stream.
-
Streams with hasAudio or hasVideo == false will not have scores calculated for those tracks, and will not take a penalty. In these cases, the only available score is presented as the overall score for the subscriber.