TL;DR
FunASR is an industrial-grade end-to-end speech recognition toolkit that supports 50+ languages, 170x real-time speed, speaker diarization, emotion detection, streaming, and OpenAI-compatible API.
Key features
High-speed inference: Up to 170x real-time processing compared to Whisper
Multilingual support: 50+ languages, including Chinese dialects
End-to-end pipeline: VAD → ASR → speaker diarization → emotion detection → punctuation in a single model call
LLM-based ASR: Fun-ASR-Nano (SenseVoice + Qwen3-0.6B) achieves top accuracy for 31 languages
Streaming support: Real-time speech recognition
Easy deployment: OpenAI-compatible API server, MCP server, vLLM acceleration, various deployment matrices
When to use it
When real-time meeting minutes generation and speaker labeling are needed
For call centers and interview analysis requiring multilingual speech recognition and emotion analysis
In production environments needing faster inference than Whisper
When integrating voice interfaces with AI agents (Claude, LangChain, Dify)