The meeting focused on the challenge of reducing latency in live streaming and the lack of industry-wide focus on the issue. Robin and Rob discussed the technical complexities, including the role of player buffers, SRT, and the difficulty of pinpointing where delays occur in the video delivery chain. They explored the idea of using AI to analyze latency by comparing inputs like SRT feeds with outputs like HLS streams, possibly using visual cues like a clock overlay. The conversation also touched on organizational questions within SVTA, considering whether latency should be its own subgroup separate from live streaming. Robin shared his skepticism about AI but noted recent improvements with curated prompts. They agreed that the concept of using AI for latency analysis was promising and worth further exploration.
Rob
- Call Robin offline after the meeting to discuss the other topic Robin pinged him about yesterday.
Robin
- Call the Netflix engineer (formerly from Comcast) to discuss ideas and get inspiration on latency and AI approaches.
- Review the Google Doc sent by Rob regarding AI analysis of latency issues.
Collaboration
- Robin and Rob: Iterate on a statement of work for a potential new SVTA subgroup focused on latency, possibly coordinating with the existing AI working group.
- Antitrust Statement & WG Discussion
Committee Direction and ATSC Latency
- Robin and Rob discussed the need to define a larger direction for their committee, noting low attendance and limited specific agenda items.
- Robin shared insights from consulting work with ATSC on latency differences between ATSC 3.0 and ATSC 1.0, explaining that while technical optimizations could yield microsecond improvements, the main latency differences stem from implementation issues in receivers and players rather than fundamental physics.
Video Latency and Decision Making
- Robin discussed a previous meeting in Burbank where a representative from ITV or Channel 4 argued that high latency was necessary for ad splicing, though Robin was initially taken aback by this position.
- Robin identified that many companies face challenges due to a lack of understanding about video latency at the management level, where performance metrics are measured, rather than by technical experts who understand the industry requirements.
- The discussion highlighted the disconnect between technical understanding and decision-making authority within companies, particularly noting that while SVTA members understand the technical aspects, the actual decision-makers at companies may not have the same expertise.
AI for Video Latency Solutions
- Robin and Rob discussed the potential use of AI to address latency issues in video systems, with Robin acknowledging that while initially skeptical, carefully curated prompts and established guardrails have reduced his reservations.
- They explored the possibility of scoping this into a proper SVTA working group project, contingent on being able to identify and analyze the sources of delay in video streams.
Streaming Latency Monitoring Challenges
- Rob and Robin discussed challenges in monitoring and measuring latency in streaming systems, particularly focusing on how to instrument the system for AI to detect delays.
- They explored potential approaches including using monitoring probes, prompt engineering, and collaborating with companies like YouTube TV or Disney Streaming to expose statistics for analysis.
- The conversation also touched on using SRT input as a reference point to measure delays between real-time content and final HLS video delivery, though Robin raised concerns about the computational cost of processing high-bitrate mezzanine compressed content for AI analysis.
AI Video Clock Analysis Feasibility
- Robin and Rob discussed the technical and financial feasibility of using AI to analyze video content, specifically focusing on identifying clocks in sports broadcasts.
- They explored the possibility of using brute force methods as a test, with Rob indicating it should be technically feasible within their credit limits.
- The conversation included ideas for optimizing the process by using prompts to ignore non-essential video information and potentially creating a simplified clock overlay.
- They also discussed the possibility of generalizing the model for broader use and the need to provide the AI with relevant documentation to address latency issues.
Latency Focus Group Discussion
- Rob and Robin discussed the need for a dedicated focus on latency, particularly in the context of live streaming and video.
- They considered coordinating with an existing AI working group in SVTA, though Robin expressed skepticism about the current quality of work.
- The conversation ended with Robin planning to consult with colleagues including Jason, Glenn, Ali, and Netflix contacts about potentially renaming the group to focus specifically on latency rather than live streaming.
AI Advisor for Video Latency
- Robin and Rob discussed developing an AI advisor to help identify latency issues in video streaming, particularly focusing on SRT latency and player buffer problems.
- They agreed this could be a promising approach, combining a real technical problem with new AI capabilities.
- The pair decided to create a statement of work for a potential working group focused on latency issues, with Robin planning to review Rob’s Google Doc on AI analysis of latency problems after the meeting.
- They also discussed the distinction between live and VOD streaming, noting that solving live streaming latency would address VOD issues as well.
Next Meeting: September 3, 2026 9 am PDT