FunASR
FunASR, from Alibaba's ModelScope community, is an industrial-grade speech recognition toolkit for offline, streaming and edge deployment. Beyond transcription it covers voice activity detection, punctuation restoration, speaker pipelines, and emotion and audio-event models, and it can serve models through an OpenAI-compatible API and MCP.
It suits developers building transcription, meeting notes, voice assistants and voice agents, particularly for Chinese. Model options include Fun-ASR-Nano (Chinese, English, Japanese and Chinese dialects and regional accents), a multilingual variant, and SenseVoiceSmall for recognition with emotion and audio-event tags, including a GGUF build for CPU and edge use with llama.cpp. A deployment matrix covers GPU, CPU, streaming servers and Docker runtimes, and the newest models can even run through Hugging Face Transformers without the toolkit.
Getting started: install PyTorch and `pip install funasr`, then load a model with `AutoModel`, for example SenseVoiceSmall with the FSMN VAD model. For serving, start the OpenAI-compatible server and send audio to `/v1/audio/transcriptions`.
Features
Why choose it
Trade-offs
Maintenance check
- Last commit
- 2026-10-22
- Commits, last 90 days
- 100+
- Releases, last 90 days
- 27
- Latest release
- v1.4.16 (2026-09-18)
- Open issues
- 53
- Recent issues with at least one comment
- 8/11
Oldest week on the left; the latest four weeks are orange. Source: GitHub commit activity.
Status rule: archived on GitHub = Archived; last commit within 90 days = Active; 91 to 365 days = Slowing; older = Stale. All figures come from the GitHub API on the date shown.
Most discussed open issues
- #3302 增加Hotwords會造成語句變短及亂取代其他文字34 comments · 2026-07-22
- #3203 建议给一个 roadmap28 comments · 2026-07-12
- #3479 [Bug] Vulkan backend crashes on AMD GPUs on Windows (0xc0000005)22 comments · 2026-08-07
- #3038 关于serve_realtime_ws.py服务的客户端测试达不到实时16 comments · 2026-06-29
- #3773 中文断句的老问题，才发现代码调用和命令行方式运行的结果好像是有差别的，不一定适用所有音频文件哈11 comments · 2026-10-14
Issue titles are quoted from GitHub. Open the link to check whether a problem still applies to the version you use.
Who it fits
- You are building transcription, meeting notes or a voice agent and need the recognition layer, especially for Chinese with dialects and accents.
- You want more than transcription: voice activity detection, punctuation, speaker pipelines and emotion tags.
- You want to deploy on GPU, CPU, streaming servers, Docker or edge through GGUF, or serve an OpenAI-compatible API.
- You want a finished assistant: it is a toolkit for the recognition layer.
- You plan commercial use without checking model licences: weights are licensed separately from the MIT code.
- Your main languages are outside Chinese and a handful of others: coverage varies by model.
How it differs
vs runanywhere-sdks★ 10.3kRunAnywhere runs Whisper speech-to-text inside mobile apps. FunASR is a server-side and edge toolkit with stronger Chinese coverage.
Compatibility
Quick start
Use cases
Alternatives
Featured in these lists
Comments
No comments yet. Be the first!