Android Gets a Voice: Building Real-Time Agents
summary
Real-time Android voice agent that listens, thinks, and speaks simultaneously using Jetpack Compose UI, a foreground audio service, streaming ST, an LLM, and low-latency TTS; three production demos: on-device GVP using MLKit gen + AI Core with Android TTS achieving ~<700 ms TTFT, cloud Cliff over WebSockets using Deepgram Flux ST + Deepgram Ora2 (Fujen) and backend ephemeral tokens at ~<800 ms, and Pipecat/WebRTC agentic Pika driving OpenClaw with Pipecat WebRTC + OpenClaw where barge‑in, echo cancellation, and the animated visualizer all work end-to-end.
AI-written summary · about
chapters (8)
- 0:00Android voice systems
- 1:42Coding with AI agents
- 4:11Voice fundamentals
- 8:32Three voice architectures
- 8:54Demo: Google Voice Pipeline
- 12:53Demo: Cliff app
- 19:18Agentic voice solutions
- 28:00Key takeaways and conclusion
related talks
42mThe State of Voice Interactions on Android
43mThe State of Voice Interactions on Android
47mBuilding Voice-First Android Apps- 47mBuilding Voice-First Android Apps
01h 07mBuilding and Scaling an AI Agent Swarm of low latency real time voice bots
35mHaving a Voice on Android: How hard can it be?
55mConversational power for Android apps
01h 58mWhat If Your Backend Could Talk? Building Real-Time AI Voice Agents