LiveKit Agents
Real-time voice agent framework built on LiveKit's WebRTC infrastructure. Handles low-latency audio streaming, voice activity detection, and multi-participant sessions. Strong for production deployments where latency and scalability are critical. 9K+ GitHub stars.
Deepgram
Real-time and batch speech recognition API with <300ms latency. Supports 30+ languages, speaker diarization, and custom vocabulary. Nova-3 model is best-in-class for English accuracy.