Making Meta RTC Audio More Resilient
summary
Meta RTC uses NACK, in-band FEC, and duplication over WebRTC and measures impact via neteq’s delay histogram and P95 latency. In experiments duplication alone keeps one-way latency low (example: 360 ms target rising to 540 ms when RTX shifts the histogram right), while NACK plus duplication recovers more audio once loss exceeds ~15% but increases jitter and buffer delay. Teams must trade extra bitrate or round trips against higher end-to-end latency and unified control across protections.
AI-written summary · about
chapters (14)
- 0:00Intro
- 0:17Agenda
- 0:52About Meta
- 1:56RTC Audio Problems
- 3:17RTC Audio Resilience
- 4:27Negative Acknowledgement NAC
- 5:53Inbound FVC
- 8:31Duplication
- 10:29Summary
- 12:34Latency
- 13:47Endtoend latency
- 14:59Net EQ
- 16:33Retransmission
- 19:54Conclusion
related talks
19mReal-time audio at Meta scale
21mOptimizing the customer experience
42mBecoming Latency-Native
17mYour realtime AI is ngmi
22mBringing RTC to Metaverse
23mVideo @Scale Fall 2022 LIVE Q&A #2
08mJackTrip WebRTC: Networked Music Experiments In A Web Browser
44mMatrixRTC: The (not anymore) missing piece to enable great video conferences