Mastering WebRTC for Large Conference Calls on Android
summary
WebRTC is usable for thousands of participants if audio and video move to the server and clients are slimmed down: VK aims for ~1,000+ attendees by using a media server to mix all incoming audio into one server-side track (clients receive one stream and metadata), asking the server for needed video resolutions and reuse a fixed small set of video tracks (e.g., ~10) instead of one per participant, achieving ~30% audio bandwidth savings, using TensorFlow/ML models and native C++ audio NN processing for noise suppression and fallbacks to H.264/VP9 tradeoffs.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 1:52WebRTC basics and architecture
- 4:42One-to-one to group calls
- 11:25Android audio processing
- 18:34Video codecs and screen sharing
- 26:16AR effects and backgrounds
- 35:16Call management APIs
- 35:39Optimizing large group calls
- 41:25Track management for scalability
- 48:58Network and data channel efficiency
- 51:05Summary
- 52:11Q&A
related talks
57mVK Calls: Connecting Thousands of People on Android and iOS via WebRTC
01h 01mCalls with 2,000+ Participants, or What WebRTC Can Do
32mWebRTC Update
53mA Million Video Calls a Day, or Call Your Mom
45mVK Calls: Raising the Bar for WebRTC Call Quality
22mWebRTC: Connecting Embedded, Mobile and Desktop with the Lowest Latencies
44mSeamless mobile real-time communication with WebRTC
36mWebRTC Use Cases
