Streaming Media
Artificial Intelligence has become a buzzword in streaming video with the promise of revolutionizing how we create, compress, and distribute videos. It’s crucial to separate the hype from reality and rare to find a vendor willing to do so. Media Excel’s CEO Narayanan Rajan is one.
article here
“I think there's pressure on vendors to satisfy certain segments of the market that you're working with things that are AI and ML related,” Rajan tells Streaming Media. “There's also a lot of pressure on customers just to implement the thing with all the buzz in order to satisfy their shareholders, their owners, their constituents.
“Some of the conversations we've had have been very much along that line. They may love the fact that we’ve got AI/ML and it checks a box for them but that solution may not be appropriate for the job they have in mind.”
Rajan stresses, “We don't dismiss the pressure that the market puts on our customers to satisfy the checklist item. It's a real thing and we respect that. We saw similar in terms of cloud transition and we're seeing it now. Some of those customers that transitioned to cloud have returned to on-prem, especially for 24x7 live streams because the cost can be prohibitive.”
Media Excel has been two decades in the business of delivering encoding and transcoding solutions for broadcast and streaming. Its most recent development HERO DIVA (Dynamic Intelligent Video Adaptive) encoding is a AI-driven software with a claimed 20% improvement on HEVC efficiency.
Quizzed on this Rajan says Media Excel has trained its algorithm on tens of thousands of hours of HEVC content and has demonstrated at least 20% efficiencies. HEVC was the priority. Now they’re training it on AVC (H.264) data. “Currently on less than 10,000 pieces of content, so we're getting somewhere in the eight to ten percent savings,” Rajan says. “But as we train and tune the model further we expect to be able to increase the level of savings we can get.”
Training DIVA on VVC is the next plan and Rajan anticipates savings with this advanced codec in the 20% range. “Until we actually do it, it's a bit of a hypothetical,” he admits.
Built into DIVA is a video analyzer that looks at the incoming video stream, compares it to what DIVA has already learned with the training data set and then applies the best set of quantization parameters to that particular piece of content.
“We support all deployment modes whether it's an appliance equipped with a GPU or it's a virtual machine or a container in the cloud. The only requirement for DIVA is to be in a GPU-instance.”
The software is a licensable feature of its core HERO 6000 platform. It is codec independent and edge device independent.
“One of the motivations for DIVA for us was to contemplate what would happen if our customers did not have to wait for the typical 10- to 12-year codec transition timeline,” Rajan says. “What if they didn’t have to wait to implement the codec into all the edge devices before the end-to-end benefit can be realized. What if we could implement a ML technology that could be quality focussed, target bit rate focused and offer quantization parameter modifications - but would not affect anything else downstream at the edge devices.”
“For those streaming at some level of scale the cost to achieve 20% bandwidth optimization would be insignificant compared to the distribution cost saving. Plus, you would not have to wait for all the edge devices to catch-up. You could operate with the devices that are out in the market already.”
Candidly he admits that DIVA comes with some computational overhead making it less suitable for customers with a smaller subscriber base.
“But if you are crossing a particular threshold of subscribers and streams there is an intersection point where it absolutely makes sense. For those customers the video infrastructure costs rapidly become irrelevant in terms of the cost savings you get on the distribution side.
“We think that even the customers who choose us initially because AI/ML is a checklist item will benefit downstream from having that feature enabled on their platform.
“There's a deeper truth in our industry which is business uncertainty drives the need to control costs. Because of that customers are always asking the question, ‘what can you do to help me control the total cost of work?’”
“We support all the different workflows including VOD and live to VOD but live streaming is the thing that's hard to do. If you get that right you will get the other stuff right as well. So that's really been a focus point for us.”
Media Excel is working on more AI developments including DIVA Pre-Process which is part of the company’s ongoing work to improve visual quality by detecting artefacts and automatically correcting for it as well as optimizing HDR.
Another piece of development is around audio detection, subtitle generation and language translation based on existing AI libraries.
“These kinds of efforts are a little bit fragmented. So while Google for example may have a great library for this kind of translation it may not have the best library for, say, Korean audio detection or Japanese translation. We want to be able to choose the best libraries, incorporate it into our product and really create something that stands on its own.”
It also plans to collaborate with other AI developers. One example is a tie up with a scene detection specialist for fast turnaround and publishing of clips and highlights in sports.
Media Excel claims its products power 400 million multi-screen subscribers worldwide. Its broadcast and streaming customers include ESPN, F1, QVC and the Home Shopping Network and Telenor. Spanish soccer league La Liga also uses Media Excel product in its streaming platform.
But despite twenty years in business the company “feels a little bit new” says Rajan, not because it lacks for product or market acceptance but because it has a “visibility problem.”
That’s partly why the former MediaKind executive was hired as interim CEO ten weeks ago – and has now made the position permanent.
“The company isn’t a MediaKind or a Harmonic with a big portfolio having to drive hundreds of millions of revenue. We're a small engineering-led company but because of the transition through various leadership modes over the last three years now it's time to get out there and let the world know what we're doing.”
Media Excel will be at IBC Show showcasing DIVA and demoing live low latency 3 seconds glass to glass with HERO 6000.
No comments:
Post a Comment