\n\n
\n Industrial speech recognition toolkit for offline, streaming, and edge deployment.
\n ASR · VAD · punctuation · speaker pipelines · emotion and audio-event models · OpenAI-compatible serving\n
\n ![\"PyPI\"](\"https://img.shields.io/pypi/v/funasr\")
![\"Stars\"](\"https://img.shields.io/github/stars/modelscope/FunASR?style=social\")
![\"Downloads\"](\"https://img.shields.io/pypi/dm/funasr\")
![\"Docs\"](\"https://img.shields.io/badge/docs-online-blue\"){kind=icon}
![\"MCP](\"https://mcptoplist.com/badge/io.github.modelscope%2Ffunasr-mcp.svg\"){kind=icon}
\n
\n\n
\n Quick Start · Colab · Benchmark · Model selection · Migration guide · Use cases · Community integrations · Deployment matrix · Deployment hub · Troubleshooting · Models · Agent Integration · OpenClaw · Docs · Contribute\n
\n\n---\n\n## Quick Start\n\n[](https://colab.research.google.com/github/modelscope/FunASR/blob/main/examples/colab/funasr_quickstart.ipynb)\n\nNo local setup? Open the [Colab quickstart](./examples/colab/) to transcribe a public sample or upload your own audio in a browser.\n\nFound FunASR useful? [Star the project](https://github.com/modelscope/FunASR) so more builders can find it.\n\n```bash\n# CPU-only installs can use the default PyPI wheels.\npip install torch torchaudio\npip install funasr\n```\n\nFor GPU quickstarts, install the PyTorch and torchaudio wheels that match your\nNVIDIA driver from [pytorch.org](https://pytorch.org/get-started/locally/)\nbefore installing FunASR. After installation, confirm the GPU is visible:\n\n```bash\npython - <<'PY'\nimport torch\nprint(torch.cuda.is_available())\nPY\n```\n\nOnly use `device=\"cuda\"` when this prints `True`; otherwise use `device=\"cpu\"`\nor reinstall PyTorch with the correct CUDA wheel.\n\n**Flagship model — Fun-ASR-Nano** (LLM-ASR for Chinese, English, and Japanese, plus Chinese dialect groups and regional accents; needs a GPU):\n\n```python\nfrom funasr import AutoModel\n\nmodel = AutoModel(model=\"FunAudioLLM/Fun-ASR-Nano-2512\", device=\"cuda\")\nresult = model.generate(input=\"https://isv-data.oss-cn-hangzhou.aliyuncs.com/ics/MaaS/ASR/test_audio/asr_example_zh.wav\")\nprint(result[0][\"text\"])\n# 欢迎大家来体验达摩院推出的语音识别模型。\n```\n\nFor the separate 31-language checkpoint, use\n[Fun-ASR-MLT-Nano-2512](https://huggingface.co/FunAudioLLM/Fun-ASR-MLT-Nano-2512).\nLanguage coverage is checkpoint-specific, so Nano and MLT-Nano should be treated as distinct model choices.\n\nOn CPU (or for five-language ASR plus emotion and audio-event tags), use\n**SenseVoiceSmall**. The pipeline below composes SenseVoiceSmall with FSMN-VAD\nand CAM++; diarization is provided by the separate CAM++ model, not by the\nSenseVoiceSmall checkpoint:\nSee the [SenseVoice paper](https://arxiv.org/abs/2407.04051),\n[Hugging Face checkpoint](https://huggingface.co/FunAudioLLM/SenseVoiceSmall),\nand [GGUF edge checkpoint](https://huggingface.co/FunAudioLLM/SenseVoiceSmall-GGUF).\n\n```python\nfrom funasr import AutoModel\nfrom funasr.utils.postprocess_utils import rich_transcription_postprocess\n\nmodel = AutoModel(model=\"iic/SenseVoiceSmall\", vad_model=\"fsmn-vad\", spk_model=\"cam++\", device=\"cuda\") # use device=\"cpu\" if you don't have a GPU\nresult = model.generate(\n input=\"https://isv-data.oss-cn-hangzhou.aliyuncs.com/ics/MaaS/ASR/test_audio/asr_example_zh.wav\",\n batch_size_s=300,\n)\n\n# The AutoModel pipeline returns VAD segments with speaker ids and timestamps:\nfor seg in result[0][\"sentence_info\"]:\n print(f\"[{seg['start']/1000:.1f}s] Speaker {seg['spk']}: {rich_transcription_postprocess(seg['sentence'])}\")\n```\n\n**Output** — structured text with speaker labels, timestamps, and punctuation:\n```\n[0.6s] Speaker 0: 欢迎大家来体验达摩院推出的语音识别模型\n```\n\nOne `AutoModel` pipeline call coordinates the configured ASR, VAD, and speaker\nmodels and returns the combined result.\n\n### Scale & deploy the flagship\n\nAt scale, accelerate Fun-ASR-Nano with vLLM (batch processing):\n\n```python\nfrom funasr.auto.auto_model_vllm import AutoModelVLLM\n\nmodel = AutoModelVLLM(model=\"FunAudioLLM/Fun-ASR-Nano-2512\", tensor_parallel_size=1)\nresults = model.generate([\"audio1.wav\", \"audio2.wav\"], language=\"auto\")\n```\n\n> **Deploy as API server:** `funasr-server --device cuda` → OpenAI-compatible endpoint at localhost:8000\n>\n> **Use with AI agents:** [MCP Server](examples/mcp_server/) for Claude/Cursor · [OpenAI API](examples/openai_api/) for LangChain/Dify/AutoGen\n>\n> **Use with voice agents:** [OpenClaw realtime plugin](integrations/openclaw/) for self-hosted Talk and Voice Call transcription\n\n### Why FunASR?\n\nWhisper is a single model; **FunASR is a toolkit** — you pick the right model\nper job: **Fun-ASR-Nano** (Chinese, English, Japanese, and Chinese dialects;\nGPU), **Fun-ASR-MLT-Nano** (31 languages), **SenseVoiceSmall** (five-language\nASR plus emotion and audio events), and **Paraformer** (low-latency streaming).\nThe table shows toolkit-level capabilities and names the model or pipeline that\nprovides each one:\n\n| | FunASR (toolkit) | Whisper | Cloud APIs |\n|---|---|---|---|\n| Top speed | **340x realtime** (Fun-ASR-Nano + vLLM) | 13x realtime | ~1x realtime |\n| Speaker ID | ✅ via VAD + CAM++ pipeline | ❌ Needs pyannote | ✅ Extra cost |\n| Emotion | ✅ via SenseVoice | ❌ | ❌ |\n| Languages | Checkpoint-specific (for example Qwen3-ASR 52, MLT-Nano 31, Nano zh/en/ja) | 57 | Varies |\n| Streaming | ✅ WebSocket (Paraformer) | ❌ | ✅ |\n| CPU viable | ✅ 17x realtime (SenseVoice) | ❌ Too slow | N/A |\n| Self-hosted | ✅ Yes (toolkit: MIT; model licenses vary) | ✅ MIT license | ❌ Cloud only |\n| Cost | Free | Free | $0.006/min+ |\n\nTrying FunASR for the first time? Use the [Colab quickstart](./examples/colab/) before setting up a local environment. Choosing a first model? Start with the [model selection guide](./docs/model_selection.md). Planning a switch from Whisper or a cloud ASR provider? Use the [migration guide](./docs/migration_from_whisper.md) and [benchmark example](./examples/migration/) to test representative audio, map features, and roll out safely.\n\n---\n\n## Installation\n\n```bash\npip install funasr\n```\n\n