Topic Voice AI
Turning Google Colab into an API Server to Run Speech-to-Speech Voice AI
This PoC turns Google Colab into a temporary WebSocket server for speech-to-speech AI. It combines faster-whisper large-v3, Gemini 2.5 Flash-Lite, VOICEVOX and Silero VAD, then improves latency through sentence buffering, streaming responses and parallel TTS generation.
What We Learned Building Voice AI with Gemini: The Major Difference Between a PoC and a Commercial Service
A Gemini Live API PoC can deliver real-time voice conversations quickly, but production introduces concurrency, quotas, observability, model lifecycle and device-specific audio issues. This article explains why the architecture moved from direct browser access to LiveKit and Vertex AI.
