Real-time audio at Meta scale
summary
Meta describes a production audio pipeline that ships across Messenger, Instagram, Facebook and WhatsApp and balances a low-compute baseline plus advanced ML models for AEC and noise suppression, uses AGC, codecs, jitter buffers and redundancy, and implements a privacy-compliant telemetry framework that automatically triages ~1% “no audio” reports into capture, playback, transport and UX causes. Concrete fixes include ignoring iOS interrupts for active calls, handling unexpected Bluetooth pairing, and tuning a stall detection with 2s recovery false positives that originally misfired during route switches.
AI-written summary · about
chapters (5)
- 0:00VoIP to metaverse audio
- 2:13Reliable real-time audio
- 6:30Echo and noise suppression
- 11:03Audio device challenges
- 17:49Immersive metaverse audio
related talks
21mMaking Meta RTC Audio More Resilient
17mYour realtime AI is ngmi
27mThe Importance of Audio Live Panel
20mMeta's AI Translations Inference
10mRTC @Scale, Future RTC Experiences Session - Live Q&A
23mVideo @Scale Fall 2022 LIVE Q&A #2
13mHow We Reduced Latency by 70% While Maintaining 99% Accuracy
50m.NET any good for Audio?