Zixi: your video delivery watchdog
Posted on May 23, 2021 by FEED Staff
Zixi’s intelligent data platform and machine learning predict failures, provide alerts on anomalies and optimise broadcast live video workflows
It’s time to realise that machine learning and AI technologies are not futuristic in media, but rather that they are essential now. AI is being actively incorporated into many types of media and entertainment applications today, from content recommendation to AI-enhanced editing and search optimisation, but little has been done for event correlation until now.
Other use cases, like predictive maintenance, quality assurance and resource optimisation, already have a decades-long track record in process industries like power generation and delivery. But the media space has been slower to apply those tools to its own workflows and delivery mechanisms. One reason for that may be M&E’s slower adoption of all digital, cloud-based infrastructures, where AI can really flourish and deliver results, but that is rapidly changing. Zixi is now using these principles to create a smarter content supply chain.
“What we’re doing in the media industry is not that dissimilar to the electric power life cycle,” explains Zixi CEO Gordon Brooks. “You have production, you have distribution and you have consumption, and with those parts of the chain, you have AI doing descriptive, predictive, prescriptive and optimised analytics. It’s done every day, all over the world. Zixi is taking these concepts from other industries and bringing them into the media workflow.”
As one of the world’s major video delivery platforms, Zixi generates over one billion telemetry data points a day, and it became clear that, if intelligently analysed, this data could generate invaluable insights for Zixi and its customers.
ZEN Master was the first Zixi product to roll out rules-based alerting, which includes everything from CPU usage on a decoder to network data or content quality data. Zixi then added machine learning to the mix, which has provided valuable insight and can then score the encoded video quality without seeing source video.
“Now we can predict the encoder quality without seeing the source video with 99.5% accuracy,” says Brooks.
Zixi is also able to generate predictions based on several other AI-generated quality scores. Quality of service, quality of transport, quality of content and quality of network data are combined to provide useful insights and quality scores that are immediately actionable by Zixi customers. There is a host of other data, such as published video player quality rankings, which can be brought into the Zixi platform to produce even deeper insights.
“Machine learning tells you things. It may not be what you want it to tell you or what you thought it was going to tell you. But you learn. It’s a journey” says Brooks.
The insights gleaned through machine learning don’t always have to be machine-learning solutions. Sometimes it’s more efficient to create an engineered rules-based solution depending on what machine-learning data has produced.
Zixi is taking proven concepts from other industries and bringing them into the media workflow
“With machine learning, you start training models and you end up with a bouquet of models – called a committee. The committee continually monitors itself and looks for improvements. When an improvement is significant, the committee promotes that particular algorithm. By nature it gets better and better as it gets more feedback.”
Brooks recalls being in conference calls with 25 people from multiple companies, working two hours a day every day for a week or more, trying to troubleshoot their non-Zixi related distribution issues. AI will help to streamline this kind of expensive and inefficient use of human talent, which could be better spent on creative solutions rather than picking through issues that AIs can analyse at scale.
Zixi is using AI to create solutions that pre-emptively address these problems before ever reaching a crisis stage. When there is an issue with the delivery or transport of content, teams can be presented with a limited set of possibilities, ranked in order of probability, which means they can take immediate action before there is an event.
“An average broadcast channel spends $500,000 a year on root cause analysis, that’s people trying to figure out what went wrong. The objective is to be able to do more and more without adding expensive human resources, and to make sure that you’re fixing issues before they happen,” concludes Brooks.
This first featured in the Spring 2021 issue of FEED magazine.