Live Q&A 2 with Speakers
summary
Live Q&A answers concrete operational details: speech activity uses 300 ms and 3 s windows for fast detection and stable decisions; end‑to‑end encryption relies on SRTP/secure frames (DTLS noted as alternate); bandwidth estimation combines delay- and loss-based BWE plus custom link‑capacity signals; group media uses simulcast with a single selective forwarder (SFU) chosen by historical latency; QUIC is under active evaluation and internal use for media transport.
AI-written summary · about
chapters (21)
- 0:00Q&A session
- 0:49AI features for older devices
- 2:38Media quality challenges
- 4:23Speech activity detection
- 5:06End-to-end encryption choices
- 6:10Bandwidth estimation and congestion
- 7:12Internet infrastructure impact
- 8:37Server-side speaker detection
- 9:39Downlink bandwidth algorithms
- 11:25SFU clusters for global calls
- 12:32New communication channels
- 13:56User experience metrics
- 14:39Audio capping accuracy metrics
- 16:26DDoS protection for WhatsApp
- 17:49Multiple web apps on device
- 19:33Video layer switching and bitrate
- 20:37Distributed speaker detection plans
- 23:05POP region for relay servers
- 25:14Codec selection in group calls
- 26:19QUIC for transport protocols
- 27:47Closing remarks
related talks
11mRTC @Scale, Resilience and Encryption Session | Live Q&A
10mRTC @Scale, Future RTC Experiences Session - Live Q&A
25mCacheable live under 200ms glass 2 glass. Is it VC or livestreaming? It is MOQ
22mWebRTC: Connecting Embedded, Mobile and Desktop with the Lowest Latencies
30mThere and back again: reinventing UDP streaming with QUIC
23m3 Second End to End Latency At Scale
17mGroup Call End-to-End Encryption and the Challenges of Encrypting Large Calls
21mReplacing WebRTC with Media over QUIC