FunASR overview and key facts (October 2026)

FunASR is an open-source speech processing toolkit maintained by Alibaba's DAMO Academy that covers transcription, streaming recognition, speaker diarization and emotion detection, with particular strength in Chinese.

The toolkit bundles models such as Fun-ASR-Nano (800 million parameters, Chinese, English, Japanese and regional accents), SenseVoiceSmall with emotion and audio event recognition, and the Paraformer series with timestamps. It deploys through a Python SDK, WebSocket streaming services, Docker, Kubernetes or vLLM, and as a single binary on CPU and edge devices via GGUF and llama.cpp. It lives in the ModelScope organization on GitHub. The GitHub API showed about 20,500 stars and a push on September 28, 2026.

• Transcribes calls in real timeStreaming ASR converts speech to text while a conversation is running. Contact centers or live captioning can work on their own servers without sending audio to a cloud API.
• Separates speakers and adds punctuationDiarization, voice activity detection and punctuation restoration produce readable transcripts with speaker turns. Meeting notes need less manual cleanup.
• Detects emotion and audio eventsSenseVoice recognizes emotions and events such as laughter or applause in five languages. Quality teams can flag calls for review based on tone.

• Optimized for Chinese; the README reports about three times lower character error rate than whisper.cpp on Chinese.
• Runs from GPU clusters down to CPU and edge devices as a single binary via GGUF.

Maintained by Alibaba's DAMO Academy in China; the models focus mainly on Chinese and English.

Check each model's license on its model card, since only the toolkit is MIT licensed. Emotion recognition on employee or customer calls can be sensitive; clarify legal basis, transparency and human review before use. Test accuracy for German or other European languages, as the main models focus on Chinese and English.

No reviews yet. Be the first to share how your team uses it.

Every review shows its verification level. We read each one before it goes live. How we check reviews