Build a real-time voice AI agent with Gemini Live API

Code for this episode → https://g.dev/cloud/voicedemo1
Gemini Live API docs → https://g.dev/cloud/gemini-live

Text-to-speech can talk but can't hear you. This voice agent listens while it speaks, stops the moment you interrupt, and calls your code to play music — built on the raw Gemini Live API.

*What you’ll learn:*
* Why live voice is different from TTS (audio-to-audio, both directions, streaming)
* The architecture and the four-step loop (open, send, receive, play)
* The three concepts that make it feel alive — voice activity detection, barge-in, and tools.

Built with the raw GenAI SDK so you see every moving piece.
Subscribe and join us for the next episode: the same app on Google Agent Development Kit (ADK).

⏱️ *Chapters:*
0:00 - Demo: a voice agent you can interrupt
0:23 - What you'll learn (3 parts)
0:45 - Text-to-speech vs Gemini Live: one direction vs two
1:13 - Audio-native: tone, pauses, and streaming
1:40 - The architecture: browser, backend, Gemini Live
2:05 - Why a WebSocket (and not a normal request)
2:31 - The core loop: open, send, receive, play
2:56 - Voice activity detection: how it knows your turn ended
3:50 - Barge-in: interrupting the agent mid-sentence
4:21 - The trick that makes interruption feel instant
4:38 - Tools: giving the agent things it can do
5:10 - The tool-latency rule: slow tool = silent agent
5:45 - Recap: the whole live voice agent
6:32 - On the next episode: rebuilding this with Google ADK

More resources:
Roadmap preview → https://g.dev/cloud/mma-roadmap
Gemini Live API docs → https://g.dev/cloud/gemini-live
Agent Development Kit (ADK) docs → https://g.dev/cloud/adk-docs

Follow the Annie online:
LinkedIn → https://goo.gle/annie-linkedin
X → https://goo.gle/annie-x

Watch more The Omni App → https://g.dev/cloud/modern-ai-app
🔔 Subscribe to Google Cloud Tech → https://goo.gle/GoogleCloudTech

#GeminiLiveAPI #VoiceAI #AIAgents

Speaker: Annie Wang
Products Mentioned: Gemini Live API, Agent Development Kit