Using Machine Learning to Enrich and Analyze Real-time Communications
summary
Amazon Chime SDK demonstrates concrete ML features: Opus padding carries ML embeddings to reconstruct lost audio without extra packets, enabling up to 1 second reconstruction per packet via neural compression and RFC 2198-compatible redundancy; transcription plus speaker embeddings power real‑time prompts (used by T‑Mobile 60,000 agents to reduce handle time); speaker ID searches scale to ~25,000 voices but currently struggle with overlapping talkers; legacy telephony audio is upsampled from 8k→48k super‑resolution for wideband playback while remaining backward compatible.
AI-written summary · about
chapters (8)
- 0:00Introduction
- 0:48Use Cases
- 1:15Overview
- 3:40Packet Loss Concealment
- 5:14Machine Learning
- 9:20TMobile
- 11:04Speaker Identification
- 15:57Summary
related talks
05mRTC @Scale, Audio ML Session
21mDeveloping Machine Learning Based Speech Enhancement Models for Teams and Skype
55mMachine Learning and AI in Java on AWS
20mReal-time ML analytics with Contact Lens for Amazon Connect
19mLLM-Enhanced Multimodal AI: Revolutionizing Audio/Video Interaction
31mPower-Up Your App With Apple’s Machine Learning Frameworks
29mEngineering Machine Learning Technologies
32mIncrease productivity and satisfaction with an intelligent contact center