mirror of
https://github.com/modelscope/FunASR.git
synced 2026-10-12 02:19:18 +00:00
docs: README 提升专业度——加入 logo、机构标识、规范表格、权威 tone
This commit is contained in:
游雁
@@ -1,109 +1,120 @@
|
||||
<p align="center">
|
||||
<h1 align="center">FunASR</h1>
|
||||
<p align="center"><b>Speech → Structured Text, in 3 Lines of Python</b></p>
|
||||
</p>
|
||||
<div align="center">
|
||||
<img src="docs/images/funasr_logo.jpg" width="400"/>
|
||||
|
||||
<p align="center">
|
||||
<a href="https://pypi.org/project/funasr/"><img src="https://img.shields.io/pypi/v/funasr?style=flat&color=2563eb" alt="PyPI"/></a>
|
||||
<a href="https://github.com/modelscope/FunASR/stargazers"><img src="https://img.shields.io/github/stars/modelscope/FunASR?style=flat&color=b45309" alt="Stars"/></a>
|
||||
<a href="https://github.com/modelscope/FunASR/blob/main/LICENSE"><img src="https://img.shields.io/badge/license-MIT-059669?style=flat" alt="License"/></a>
|
||||
<a href="https://modelscope.github.io/FunASR"><img src="https://img.shields.io/badge/docs-online-blue?style=flat" alt="Docs"/></a>
|
||||
</p>
|
||||
**A Fundamental End-to-End Speech Recognition Toolkit**
|
||||
|
||||
<p align="center">
|
||||
<a href="./README_zh.md">简体中文</a> ·
|
||||
<a href="https://modelscope.github.io/FunASR"><b>Documentation</b></a> ·
|
||||
<a href="https://modelscope.github.io/FunASR/tutorial.html">Tutorial</a> ·
|
||||
<a href="https://modelscope.github.io/FunASR/training.html">Training</a> ·
|
||||
<a href="https://modelscope.github.io/FunASR/api.html">API</a> ·
|
||||
<a href="https://arxiv.org/abs/2305.11013">Paper</a>
|
||||
</p>
|
||||
[](https://pypi.org/project/funasr/)
|
||||
[](https://github.com/modelscope/FunASR/stargazers)
|
||||
[](https://github.com/modelscope/FunASR/blob/main/LICENSE)
|
||||
[](https://pypi.org/project/funasr/)
|
||||
|
||||
[English](./README.md) | [简体中文](./README_zh.md) | [Documentation](https://modelscope.github.io/FunASR) | [Paper](https://arxiv.org/abs/2305.11013)
|
||||
|
||||
</div>
|
||||
|
||||
---
|
||||
|
||||
FunASR is an industrial-grade speech recognition toolkit developed by [Tongyi Lab, Alibaba Group](https://tongyi.aliyun.com). It provides a unified Python interface for the complete speech understanding pipeline — from raw audio to speaker-attributed, punctuated transcripts — in a single API call.
|
||||
|
||||
```python
|
||||
from funasr import AutoModel
|
||||
|
||||
model = AutoModel(model="paraformer-zh", vad_model="fsmn-vad",
|
||||
punc_model="ct-punc", spk_model="cam++")
|
||||
model = AutoModel(
|
||||
model="paraformer-zh", # speech recognition
|
||||
vad_model="fsmn-vad", # voice activity detection
|
||||
punc_model="ct-punc", # punctuation restoration
|
||||
spk_model="cam++", # speaker diarization
|
||||
)
|
||||
res = model.generate(input="meeting.wav", batch_size_s=300)
|
||||
|
||||
res = model.generate(input="meeting.wav")
|
||||
```
|
||||
```
|
||||
[Speaker 0] 欢迎大家来体验达摩院推出的语音识别模型。
|
||||
[Speaker 1] 非常感谢，今天我们主要讨论三个议题。
|
||||
[Speaker 0] 好的，请开始吧。
|
||||
for s in res[0]["sentence_info"]:
|
||||
print(f"[Speaker {s['spk']}] {s['text']}")
|
||||
```
|
||||
|
||||
One API call. Long audio in, speaker-labeled sentences out. VAD segmentation, speech recognition, punctuation, and speaker diarization — handled.
|
||||
**Output:**
|
||||
|
||||
> \[Speaker 0\] 欢迎大家来体验达摩院推出的语音识别模型。
|
||||
> \[Speaker 1\] 非常感谢，今天我们主要讨论三个议题。
|
||||
> \[Speaker 0\] 好的，请开始吧。
|
||||
|
||||
---
|
||||
|
||||
## Install
|
||||
## Key Features
|
||||
|
||||
- **Unified Pipeline** — VAD, ASR, punctuation, speaker diarization, emotion detection composed in one `AutoModel()` call
|
||||
- **50+ Languages** — Fun-ASR-Nano (31 languages incl. Chinese dialects), Qwen3-ASR (52 languages with auto-detection)
|
||||
- **High Performance** — Non-autoregressive inference; SenseVoice achieves 70ms RTF for 10s audio (15× faster than Whisper)
|
||||
- **Speaker Diarization** — Per-sentence speaker labels, compatible with Paraformer, Fun-ASR-Nano, and SenseVoice
|
||||
- **Emotion & Audio Events** — SenseVoice classifies emotions (happy/sad/angry/neutral) and detects BGM, laughter, applause
|
||||
- **Production Ready** — Fine-tune with DeepSpeed, export to ONNX, deploy via Docker runtime or Python SDK
|
||||
|
||||
## Installation
|
||||
|
||||
```bash
|
||||
pip install funasr
|
||||
pip install -U funasr
|
||||
|
||||
# Or install from source for latest models
|
||||
git clone https://github.com/modelscope/FunASR.git && pip install -e ./FunASR
|
||||
```
|
||||
|
||||
> **China users**: models auto-download from ModelScope. International users: add `hub="hf"` for HuggingFace.
|
||||
> Models auto-download from [ModelScope](https://modelscope.cn) (fast in China). Add `hub="hf"` for [HuggingFace](https://huggingface.co/FunASR).
|
||||
|
||||
## Why FunASR
|
||||
## Model Zoo
|
||||
|
||||
| | |
|
||||
|---|---|
|
||||
| **50+ languages** | Fun-ASR-Nano covers 31 languages; Qwen3-ASR covers 52 with auto-detection |
|
||||
| **15× faster than Whisper** | Non-autoregressive architecture; SenseVoice processes 10s audio in 70ms |
|
||||
| **Speaker diarization** | "Who said what" — works with Paraformer, Fun-ASR-Nano, and SenseVoice |
|
||||
| **Emotion & audio events** | SenseVoice detects happy/sad/angry + BGM, laughter, applause |
|
||||
| **Full lifecycle** | Fine-tune with DeepSpeed → export ONNX → deploy with Docker |
|
||||
| Model | Type | Languages | Params | Inference Speed | Links |
|
||||
|:------|:-----|:---------:|:------:|:---------------:|:-----:|
|
||||
| [Fun-ASR-Nano](https://github.com/FunAudioLLM/Fun-ASR) | ASR | 31 | 800M | 中 | [MS](https://modelscope.cn/models/FunAudioLLM/Fun-ASR-Nano-2512) · [HF](https://huggingface.co/FunAudioLLM/Fun-ASR-Nano-2512) |
|
||||
| [Paraformer-zh](https://huggingface.co/funasr/paraformer-zh) | ASR | 2 | 220M | 快 | [MS](https://modelscope.cn/models/damo/speech_paraformer-large-vad-punc_asr_nat-zh-cn-16k-common-vocab8404-pytorch) · [HF](https://huggingface.co/funasr/paraformer-zh) |
|
||||
| [SenseVoice](https://github.com/FunAudioLLM/SenseVoice) | ASR+SER+AED | 5 | 234M | 极快 (70ms/10s) | [MS](https://modelscope.cn/models/iic/SenseVoiceSmall) · [HF](https://huggingface.co/FunAudioLLM/SenseVoiceSmall) |
|
||||
| [Qwen3-ASR](https://huggingface.co/Qwen/Qwen3-ASR-1.7B) | ASR (LLM) | 52 | 1.7B | 慢 | [MS](https://modelscope.cn/models/Qwen/Qwen3-ASR-1.7B) · [HF](https://huggingface.co/Qwen/Qwen3-ASR-1.7B) |
|
||||
| [GLM-ASR-Nano](https://huggingface.co/zai-org/GLM-ASR-Nano-2512) | ASR (LLM) | 17 | 1.5B | 慢 | [MS](https://modelscope.cn/models/ZhipuAI/GLM-ASR-Nano-2512) · [HF](https://huggingface.co/zai-org/GLM-ASR-Nano-2512) |
|
||||
| Paraformer-Streaming | ASR | 1 | 220M | 实时 | [MS](https://modelscope.cn/models/damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-online) · [HF](https://huggingface.co/funasr/paraformer-zh-streaming) |
|
||||
| fsmn-vad | VAD | 2 | 0.4M | — | [MS](https://modelscope.cn/models/damo/speech_fsmn_vad_zh-cn-16k-common-pytorch) · [HF](https://huggingface.co/funasr/fsmn-vad) |
|
||||
| ct-punc | Punctuation | 2 | 290M | — | [MS](https://modelscope.cn/models/damo/punc_ct-transformer_cn-en-common-vocab471067-large) · [HF](https://huggingface.co/funasr/ct-punc) |
|
||||
| cam++ | Speaker | — | 7.2M | — | [MS](https://modelscope.cn/models/iic/speech_campplus_sv_zh-cn_16k-common) · [HF](https://huggingface.co/funasr/campplus) |
|
||||
| emotion2vec+ | Emotion | — | 300M | — | [MS](https://modelscope.cn/models/iic/emotion2vec_plus_large) · [HF](https://huggingface.co/emotion2vec/emotion2vec_plus_large) |
|
||||
|
||||
## Models
|
||||
|
||||
| Model | Languages | Speed | Best for | Links |
|
||||
|:------|:---------:|:-----:|:---------|:-----:|
|
||||
| **[Fun-ASR-Nano](https://github.com/FunAudioLLM/Fun-ASR)** | 31 | ●●●○ | Multi-language, dialects, lyrics | [ModelScope](https://modelscope.cn/models/FunAudioLLM/Fun-ASR-Nano-2512) · [HF](https://huggingface.co/FunAudioLLM/Fun-ASR-Nano-2512) |
|
||||
| **Paraformer** | 2 | ●●●● | Chinese production ASR | [ModelScope](https://modelscope.cn/models/damo/speech_paraformer-large-vad-punc_asr_nat-zh-cn-16k-common-vocab8404-pytorch) · [HF](https://huggingface.co/funasr/paraformer-zh) |
|
||||
| **SenseVoice** | 5 | ●●●●● | ASR + emotion + audio events | [ModelScope](https://modelscope.cn/models/iic/SenseVoiceSmall) · [HF](https://huggingface.co/FunAudioLLM/SenseVoiceSmall) |
|
||||
| **[Qwen3-ASR](https://huggingface.co/Qwen/Qwen3-ASR-1.7B)** | 52 | ●●○○ | Highest accuracy, context-aware | [ModelScope](https://modelscope.cn/models/Qwen/Qwen3-ASR-1.7B) · [HF](https://huggingface.co/Qwen/Qwen3-ASR-1.7B) |
|
||||
| **Paraformer-Streaming** | 1 | real-time | Live transcription | [ModelScope](https://modelscope.cn/models/damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-online) · [HF](https://huggingface.co/funasr/paraformer-zh-streaming) |
|
||||
| **cam++** | — | — | Speaker diarization | [ModelScope](https://modelscope.cn/models/iic/speech_campplus_sv_zh-cn_16k-common) · [HF](https://huggingface.co/funasr/campplus) |
|
||||
| **emotion2vec+** | — | — | Speech emotion recognition | [ModelScope](https://modelscope.cn/models/iic/emotion2vec_plus_large) · [HF](https://huggingface.co/emotion2vec/emotion2vec_plus_large) |
|
||||
|
||||
Speed: ●●●●● = ultra-fast (70ms/10s audio), ●●○○ = slower (LLM-based). Full list → [model_zoo/](./model_zoo)
|
||||
Full model list → [model_zoo/](./model_zoo) | Detailed usage → [Documentation](https://modelscope.github.io/FunASR/tutorial.html)
|
||||
|
||||
## What's New
|
||||
|
||||
- **2026.05** — [Qwen3-ASR](https://huggingface.co/Qwen/Qwen3-ASR-1.7B): 52-language LLM-based ASR with auto language detection
|
||||
- **2026.05** — [GLM-ASR-Nano](https://huggingface.co/zai-org/GLM-ASR-Nano-2512): 17-language robust ASR with dialect optimization
|
||||
- **2026.05** — Speaker diarization support for Fun-ASR-Nano and SenseVoice
|
||||
- **2025.12** — [Fun-ASR-Nano](https://github.com/FunAudioLLM/Fun-ASR): 31-language ASR trained on tens of millions of hours
|
||||
- **2024.07** — [SenseVoice](https://github.com/FunAudioLLM/SenseVoice): multi-task speech understanding (ASR + emotion + events)
|
||||
| Date | Update |
|
||||
|:-----|:-------|
|
||||
| 2026.05 | [Qwen3-ASR](https://huggingface.co/Qwen/Qwen3-ASR-1.7B) — 52 languages, LLM-based, auto language detection |
|
||||
| 2026.05 | [GLM-ASR-Nano](https://huggingface.co/zai-org/GLM-ASR-Nano-2512) — 17 languages, dialect & low-volume optimization |
|
||||
| 2026.05 | Speaker diarization support added for Fun-ASR-Nano and SenseVoice |
|
||||
| 2025.12 | [Fun-ASR-Nano](https://github.com/FunAudioLLM/Fun-ASR) — 31-language end-to-end ASR, trained on tens of millions of hours |
|
||||
| 2024.07 | [SenseVoice](https://github.com/FunAudioLLM/SenseVoice) — multi-task speech understanding (ASR + emotion + events) |
|
||||
|
||||
<details><summary>Earlier releases</summary>
|
||||
<details><summary>View full changelog</summary>
|
||||
|
||||
- 2024.10 — Whisper-large-v3-turbo support
|
||||
- 2024.09 — Keyword spotting models
|
||||
- 2024.03 — Qwen-Audio, Whisper-large-v3 support
|
||||
- 2024.01 — FunASR 1.0 released
|
||||
- 2023.11 — File transcription service 3.0 (CPU)
|
||||
- 2023.10 — Paraformer-VAD-SPK combined pipeline
|
||||
- 2023.08 — Real-time transcription service
|
||||
See [CHANGELOG.md](./CHANGELOG.md) for complete release history.
|
||||
|
||||
</details>
|
||||
|
||||
## Learn More
|
||||
|
||||
| Resource | Description |
|
||||
|:---------|:------------|
|
||||
| [Tutorial](https://modelscope.github.io/FunASR/tutorial.html) | Install, choose a model, run ASR/VAD/speaker diarization |
|
||||
| [Training Guide](https://modelscope.github.io/FunASR/training.html) | Fine-tune Paraformer, SenseVoice, Fun-ASR-Nano on custom data |
|
||||
| [Developer Guide](https://modelscope.github.io/FunASR/model-registration.html) | Add a new model, understand the registry, test & contribute |
|
||||
| [API Reference](https://modelscope.github.io/FunASR/api.html) | Auto-generated class & method docs with source links |
|
||||
| [Runtime / Deployment](./runtime/readme.md) | File transcription service, real-time streaming service (CPU/GPU) |
|
||||
|
||||
## Ecosystem
|
||||
|
||||
| | |
|
||||
|:--|:--|
|
||||
| [**Fun-ASR-Nano**](https://github.com/FunAudioLLM/Fun-ASR) | Multi-language ASR large model — 31 languages, timestamps, hotwords |
|
||||
| [**SenseVoice**](https://github.com/FunAudioLLM/SenseVoice) | Multi-task speech understanding — ASR, emotion, audio events |
|
||||
| [**FunClip**](https://github.com/modelscope/FunClip) | AI video clipping powered by FunASR + LLM editing |
|
||||
| [**CosyVoice**](https://github.com/FunAudioLLM/CosyVoice) | Natural speech generation with voice cloning |
|
||||
| Project | Description |
|
||||
|:--------|:------------|
|
||||
| [Fun-ASR-Nano](https://github.com/FunAudioLLM/Fun-ASR) | Multi-language ASR large model — 31 languages, timestamps, hotwords, speaker diarization |
|
||||
| [SenseVoice](https://github.com/FunAudioLLM/SenseVoice) | Multi-task speech understanding — ASR, language ID, emotion, audio events |
|
||||
| [FunClip](https://github.com/modelscope/FunClip) | AI video clipping powered by FunASR and LLM-assisted editing |
|
||||
| [CosyVoice](https://github.com/FunAudioLLM/CosyVoice) | Natural speech generation with multi-language, timbre, and emotion control |
|
||||
|
||||
## Community
|
||||
|
||||
Questions → [GitHub Issues](https://github.com/modelscope/FunASR/issues)
|
||||
Discussion → DingTalk group (scan below)
|
||||
- Issues & feature requests → [GitHub Issues](https://github.com/modelscope/FunASR/issues)
|
||||
- Join our DingTalk discussion group:
|
||||
|
||||
<img src="docs/images/dingding.png" width="150"/>
|
||||
|
||||
@@ -111,13 +122,13 @@ Discussion → DingTalk group (scan below)
|
||||
|
||||
```bibtex
|
||||
@inproceedings{gao2023funasr,
|
||||
author={Zhifu Gao and Zerui Li and Jiaming Wang and Haoneng Luo and Xian Shi and Mengzhe Chen and Yabin Li and Lingyun Zuo and Zhihao Du and Zhangyu Xiao and Shiliang Zhang},
|
||||
title={FunASR: A Fundamental End-to-End Speech Recognition Toolkit},
|
||||
year={2023},
|
||||
booktitle={INTERSPEECH},
|
||||
author = {Zhifu Gao and Zerui Li and Jiaming Wang and Haoneng Luo and Xian Shi and Mengzhe Chen and Yabin Li and Lingyun Zuo and Zhihao Du and Zhangyu Xiao and Shiliang Zhang},
|
||||
title = {FunASR: A Fundamental End-to-End Speech Recognition Toolkit},
|
||||
booktitle = {INTERSPEECH},
|
||||
year = {2023},
|
||||
}
|
||||
```
|
||||
|
||||
## License
|
||||
|
||||
Code: [MIT](./LICENSE) · Model weights: [FunASR Model License](./MODEL_LICENSE)
|
||||
Code: [MIT License](./LICENSE) · Model weights: [FunASR Model License](./MODEL_LICENSE) (commercial use permitted with attribution)
|
||||
|
||||
+86
-75
@@ -1,109 +1,120 @@
|
||||
<p align="center">
|
||||
<h1 align="center">FunASR</h1>
|
||||
<p align="center"><b>语音 → 结构化文本，3 行 Python 搞定</b></p>
|
||||
</p>
|
||||
<div align="center">
|
||||
<img src="docs/images/funasr_logo.jpg" width="400"/>
|
||||
|
||||
<p align="center">
|
||||
<a href="https://pypi.org/project/funasr/"><img src="https://img.shields.io/pypi/v/funasr?style=flat&color=2563eb" alt="PyPI"/></a>
|
||||
<a href="https://github.com/modelscope/FunASR/stargazers"><img src="https://img.shields.io/github/stars/modelscope/FunASR?style=flat&color=b45309" alt="Stars"/></a>
|
||||
<a href="https://github.com/modelscope/FunASR/blob/main/LICENSE"><img src="https://img.shields.io/badge/license-MIT-059669?style=flat" alt="License"/></a>
|
||||
<a href="https://modelscope.github.io/FunASR/zh/"><img src="https://img.shields.io/badge/文档-在线-blue?style=flat" alt="Docs"/></a>
|
||||
</p>
|
||||
**基础端到端语音识别工具包**
|
||||
|
||||
<p align="center">
|
||||
<a href="./README.md">English</a> ·
|
||||
<a href="https://modelscope.github.io/FunASR/zh/"><b>完整文档</b></a> ·
|
||||
<a href="https://modelscope.github.io/FunASR/zh/tutorial.html">使用教程</a> ·
|
||||
<a href="https://modelscope.github.io/FunASR/zh/training.html">训练指南</a> ·
|
||||
<a href="https://modelscope.github.io/FunASR/api.html">API</a> ·
|
||||
<a href="https://arxiv.org/abs/2305.11013">论文</a>
|
||||
</p>
|
||||
[](https://pypi.org/project/funasr/)
|
||||
[](https://github.com/modelscope/FunASR/stargazers)
|
||||
[](https://github.com/modelscope/FunASR/blob/main/LICENSE)
|
||||
[](https://pypi.org/project/funasr/)
|
||||
|
||||
[English](./README.md) | [简体中文](./README_zh.md) | [完整文档](https://modelscope.github.io/FunASR/zh/) | [论文](https://arxiv.org/abs/2305.11013)
|
||||
|
||||
</div>
|
||||
|
||||
---
|
||||
|
||||
FunASR 是[阿里巴巴通义实验室](https://tongyi.aliyun.com)开发的工业级语音识别工具包，提供统一的 Python 接口，一次调用即可完成从原始音频到带说话人标注、标点恢复的结构化文本全流程。
|
||||
|
||||
```python
|
||||
from funasr import AutoModel
|
||||
|
||||
model = AutoModel(model="paraformer-zh", vad_model="fsmn-vad",
|
||||
punc_model="ct-punc", spk_model="cam++")
|
||||
model = AutoModel(
|
||||
model="paraformer-zh", # 语音识别
|
||||
vad_model="fsmn-vad", # 语音端点检测
|
||||
punc_model="ct-punc", # 标点恢复
|
||||
spk_model="cam++", # 说话人分离
|
||||
)
|
||||
res = model.generate(input="meeting.wav", batch_size_s=300)
|
||||
|
||||
res = model.generate(input="meeting.wav")
|
||||
```
|
||||
```
|
||||
[说话人 0] 欢迎大家来体验达摩院推出的语音识别模型。
|
||||
[说话人 1] 非常感谢，今天我们主要讨论三个议题。
|
||||
[说话人 0] 好的，请开始吧。
|
||||
for s in res[0]["sentence_info"]:
|
||||
print(f"[说话人 {s['spk']}] {s['text']}")
|
||||
```
|
||||
|
||||
一次调用，长音频进，带说话人标签的逐句结果出。VAD 分段、语音识别、标点恢复、说话人分离——全部搞定。
|
||||
**输出：**
|
||||
|
||||
> \[说话人 0\] 欢迎大家来体验达摩院推出的语音识别模型。
|
||||
> \[说话人 1\] 非常感谢，今天我们主要讨论三个议题。
|
||||
> \[说话人 0\] 好的，请开始吧。
|
||||
|
||||
---
|
||||
|
||||
## 核心特性
|
||||
|
||||
- **统一流水线** — VAD、ASR、标点、说话人分离、情感检测，一个 `AutoModel()` 调用组合完成
|
||||
- **50+ 语言** — Fun-ASR-Nano（31 种语言含方言）、Qwen3-ASR（52 种语言自动检测）
|
||||
- **高性能** — 非自回归推理；SenseVoice 处理 10 秒音频仅需 70ms（比 Whisper 快 15 倍）
|
||||
- **说话人分离** — 逐句话者标注，兼容 Paraformer、Fun-ASR-Nano、SenseVoice
|
||||
- **情感与音频事件** — SenseVoice 识别情感（开心/悲伤/愤怒/中性）并检测背景音乐、笑声、掌声
|
||||
- **生产就绪** — DeepSpeed 微调、ONNX 导出、Docker 运行时或 Python SDK 部署
|
||||
|
||||
## 安装
|
||||
|
||||
```bash
|
||||
pip install funasr
|
||||
pip install -U funasr
|
||||
|
||||
# 或从源码安装（获取最新模型支持）
|
||||
git clone https://github.com/modelscope/FunASR.git && pip install -e ./FunASR
|
||||
```
|
||||
|
||||
> 模型默认从 ModelScope 下载（国内快速）。海外用户加 `hub="hf"` 切换 HuggingFace。
|
||||
|
||||
## 为什么选 FunASR
|
||||
|
||||
| | |
|
||||
|---|---|
|
||||
| **50+ 语言** | Fun-ASR-Nano 覆盖 31 种语言含方言；Qwen3-ASR 覆盖 52 种语言自动检测 |
|
||||
| **比 Whisper 快 15 倍** | 非自回归架构；SenseVoice 处理 10 秒音频仅需 70ms |
|
||||
| **说话人分离** | "谁说了什么"——Paraformer、Fun-ASR-Nano、SenseVoice 三大模型均支持 |
|
||||
| **情感与音频事件** | SenseVoice 检测开心/悲伤/愤怒 + 背景音乐、笑声、掌声 |
|
||||
| **完整链路** | DeepSpeed 微调 → ONNX 导出 → Docker 部署 |
|
||||
> 模型默认从 [ModelScope](https://modelscope.cn) 下载（国内快速）。海外用户添加 `hub="hf"` 切换 [HuggingFace](https://huggingface.co/FunASR)。
|
||||
|
||||
## 模型
|
||||
|
||||
| 模型 | 语言 | 速度 | 适用场景 | 链接 |
|
||||
|:-----|:---:|:---:|:---------|:---:|
|
||||
| **[Fun-ASR-Nano](https://github.com/FunAudioLLM/Fun-ASR)** | 31 | ●●●○ | 多语言、方言、歌词 | [ModelScope](https://modelscope.cn/models/FunAudioLLM/Fun-ASR-Nano-2512) · [HF](https://huggingface.co/FunAudioLLM/Fun-ASR-Nano-2512) |
|
||||
| **Paraformer** | 2 | ●●●● | 中文生产级识别 | [ModelScope](https://modelscope.cn/models/damo/speech_paraformer-large-vad-punc_asr_nat-zh-cn-16k-common-vocab8404-pytorch) · [HF](https://huggingface.co/funasr/paraformer-zh) |
|
||||
| **SenseVoice** | 5 | ●●●●● | 识别 + 情感 + 音频事件 | [ModelScope](https://modelscope.cn/models/iic/SenseVoiceSmall) · [HF](https://huggingface.co/FunAudioLLM/SenseVoiceSmall) |
|
||||
| **[Qwen3-ASR](https://huggingface.co/Qwen/Qwen3-ASR-1.7B)** | 52 | ●●○○ | 最高精度，上下文理解 | [ModelScope](https://modelscope.cn/models/Qwen/Qwen3-ASR-1.7B) · [HF](https://huggingface.co/Qwen/Qwen3-ASR-1.7B) |
|
||||
| **Paraformer-Streaming** | 1 | 实时 | 实时转写 | [ModelScope](https://modelscope.cn/models/damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-online) · [HF](https://huggingface.co/funasr/paraformer-zh-streaming) |
|
||||
| **cam++** | — | — | 说话人分离 | [ModelScope](https://modelscope.cn/models/iic/speech_campplus_sv_zh-cn_16k-common) · [HF](https://huggingface.co/funasr/campplus) |
|
||||
| **emotion2vec+** | — | — | 语音情感识别 | [ModelScope](https://modelscope.cn/models/iic/emotion2vec_plus_large) · [HF](https://huggingface.co/emotion2vec/emotion2vec_plus_large) |
|
||||
| 模型 | 类型 | 语言 | 参数量 | 推理速度 | 链接 |
|
||||
|:-----|:-----|:---:|:------:|:--------:|:----:|
|
||||
| [Fun-ASR-Nano](https://github.com/FunAudioLLM/Fun-ASR) | ASR | 31 | 800M | 中 | [MS](https://modelscope.cn/models/FunAudioLLM/Fun-ASR-Nano-2512) · [HF](https://huggingface.co/FunAudioLLM/Fun-ASR-Nano-2512) |
|
||||
| [Paraformer-zh](https://huggingface.co/funasr/paraformer-zh) | ASR | 2 | 220M | 快 | [MS](https://modelscope.cn/models/damo/speech_paraformer-large-vad-punc_asr_nat-zh-cn-16k-common-vocab8404-pytorch) · [HF](https://huggingface.co/funasr/paraformer-zh) |
|
||||
| [SenseVoice](https://github.com/FunAudioLLM/SenseVoice) | ASR+情感+事件 | 5 | 234M | 极快 (70ms/10s) | [MS](https://modelscope.cn/models/iic/SenseVoiceSmall) · [HF](https://huggingface.co/FunAudioLLM/SenseVoiceSmall) |
|
||||
| [Qwen3-ASR](https://huggingface.co/Qwen/Qwen3-ASR-1.7B) | ASR (大模型) | 52 | 1.7B | 慢 | [MS](https://modelscope.cn/models/Qwen/Qwen3-ASR-1.7B) · [HF](https://huggingface.co/Qwen/Qwen3-ASR-1.7B) |
|
||||
| [GLM-ASR-Nano](https://huggingface.co/zai-org/GLM-ASR-Nano-2512) | ASR (大模型) | 17 | 1.5B | 慢 | [MS](https://modelscope.cn/models/ZhipuAI/GLM-ASR-Nano-2512) · [HF](https://huggingface.co/zai-org/GLM-ASR-Nano-2512) |
|
||||
| Paraformer-Streaming | ASR | 1 | 220M | 实时 | [MS](https://modelscope.cn/models/damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-online) · [HF](https://huggingface.co/funasr/paraformer-zh-streaming) |
|
||||
| fsmn-vad | VAD | 2 | 0.4M | — | [MS](https://modelscope.cn/models/damo/speech_fsmn_vad_zh-cn-16k-common-pytorch) · [HF](https://huggingface.co/funasr/fsmn-vad) |
|
||||
| ct-punc | 标点恢复 | 2 | 290M | — | [MS](https://modelscope.cn/models/damo/punc_ct-transformer_cn-en-common-vocab471067-large) · [HF](https://huggingface.co/funasr/ct-punc) |
|
||||
| cam++ | 说话人 | — | 7.2M | — | [MS](https://modelscope.cn/models/iic/speech_campplus_sv_zh-cn_16k-common) · [HF](https://huggingface.co/funasr/campplus) |
|
||||
| emotion2vec+ | 情感 | — | 300M | — | [MS](https://modelscope.cn/models/iic/emotion2vec_plus_large) · [HF](https://huggingface.co/emotion2vec/emotion2vec_plus_large) |
|
||||
|
||||
速度：●●●●● = 极快（10s 音频 70ms），●●○○ = 较慢（大模型）。完整列表 → [model_zoo/](./model_zoo)
|
||||
完整模型列表 → [model_zoo/](./model_zoo) | 详细用法 → [使用教程](https://modelscope.github.io/FunASR/zh/tutorial.html)
|
||||
|
||||
## 最新动态
|
||||
|
||||
- **2026.05** — [Qwen3-ASR](https://huggingface.co/Qwen/Qwen3-ASR-1.7B)：52 种语言，基于大模型，自动语种检测
|
||||
- **2026.05** — [GLM-ASR-Nano](https://huggingface.co/zai-org/GLM-ASR-Nano-2512)：17 种语言，方言和低音量优化
|
||||
- **2026.05** — Fun-ASR-Nano 和 SenseVoice 新增说话人分离支持
|
||||
- **2025.12** — [Fun-ASR-Nano](https://github.com/FunAudioLLM/Fun-ASR)：31 种语言，数千万小时数据训练
|
||||
- **2024.07** — [SenseVoice](https://github.com/FunAudioLLM/SenseVoice)：多任务语音理解（识别 + 情感 + 事件）
|
||||
| 时间 | 更新 |
|
||||
|:-----|:-----|
|
||||
| 2026.05 | [Qwen3-ASR](https://huggingface.co/Qwen/Qwen3-ASR-1.7B) — 52 种语言，基于大模型，自动语种检测 |
|
||||
| 2026.05 | [GLM-ASR-Nano](https://huggingface.co/zai-org/GLM-ASR-Nano-2512) — 17 种语言，方言和低音量优化 |
|
||||
| 2026.05 | Fun-ASR-Nano 和 SenseVoice 新增说话人分离支持 |
|
||||
| 2025.12 | [Fun-ASR-Nano](https://github.com/FunAudioLLM/Fun-ASR) — 31 种语言，数千万小时数据训练 |
|
||||
| 2024.07 | [SenseVoice](https://github.com/FunAudioLLM/SenseVoice) — 多任务语音理解（识别 + 情感 + 事件） |
|
||||
|
||||
<details><summary>更早版本</summary>
|
||||
<details><summary>查看完整更新历史</summary>
|
||||
|
||||
- 2024.10 — 新增 Whisper-large-v3-turbo 支持
|
||||
- 2024.09 — 新增语音唤醒模型
|
||||
- 2024.03 — Qwen-Audio、Whisper-large-v3 支持
|
||||
- 2024.01 — FunASR 1.0 发布
|
||||
- 2023.11 — 离线文件转写服务 3.0（CPU）
|
||||
- 2023.10 — Paraformer-VAD-SPK 组合流水线
|
||||
- 2023.08 — 实时转写服务发布
|
||||
参见 [CHANGELOG.md](./CHANGELOG.md)。
|
||||
|
||||
</details>
|
||||
|
||||
## 了解更多
|
||||
|
||||
| 资源 | 说明 |
|
||||
|:-----|:-----|
|
||||
| [使用教程](https://modelscope.github.io/FunASR/zh/tutorial.html) | 安装、选择模型、运行识别/VAD/说话人分离 |
|
||||
| [训练指南](https://modelscope.github.io/FunASR/zh/training.html) | 在自定义数据上微调 Paraformer、SenseVoice、Fun-ASR-Nano |
|
||||
| [开发指南](https://modelscope.github.io/FunASR/zh/model-registration.html) | 添加新模型、理解注册表机制、测试与贡献 |
|
||||
| [API 文档](https://modelscope.github.io/FunASR/api.html) | 自动生成的类和方法文档，附源码链接 |
|
||||
| [部署文档](./runtime/readme_cn.md) | 离线文件转写服务、实时流式服务（CPU/GPU） |
|
||||
|
||||
## 生态
|
||||
|
||||
| | |
|
||||
|:--|:--|
|
||||
| [**Fun-ASR-Nano**](https://github.com/FunAudioLLM/Fun-ASR) | 多语言语音识别大模型——31 种语言、时间戳、热词 |
|
||||
| [**SenseVoice**](https://github.com/FunAudioLLM/SenseVoice) | 多任务语音理解——识别、情感、音频事件 |
|
||||
| [**FunClip**](https://github.com/modelscope/FunClip) | 基于 FunASR + 大模型的 AI 视频剪辑 |
|
||||
| [**CosyVoice**](https://github.com/FunAudioLLM/CosyVoice) | 自然语音生成，支持音色克隆 |
|
||||
| 项目 | 说明 |
|
||||
|:-----|:-----|
|
||||
| [Fun-ASR-Nano](https://github.com/FunAudioLLM/Fun-ASR) | 多语言语音识别大模型——31 种语言、时间戳、热词、说话人分离 |
|
||||
| [SenseVoice](https://github.com/FunAudioLLM/SenseVoice) | 多任务语音理解——识别、语种识别、情感、音频事件 |
|
||||
| [FunClip](https://github.com/modelscope/FunClip) | 基于 FunASR 和大模型辅助编辑的 AI 视频剪辑 |
|
||||
| [CosyVoice](https://github.com/FunAudioLLM/CosyVoice) | 自然语音生成，支持多语言、音色和情感控制 |
|
||||
|
||||
## 社区
|
||||
|
||||
问题反馈 → [GitHub Issues](https://github.com/modelscope/FunASR/issues)
|
||||
交流讨论 → 钉钉群（扫码加入）
|
||||
- 问题反馈 → [GitHub Issues](https://github.com/modelscope/FunASR/issues)
|
||||
- 加入钉钉交流群：
|
||||
|
||||
<img src="docs/images/dingding.png" width="150"/>
|
||||
|
||||
@@ -111,13 +122,13 @@ pip install funasr
|
||||
|
||||
```bibtex
|
||||
@inproceedings{gao2023funasr,
|
||||
author={Zhifu Gao and Zerui Li and Jiaming Wang and Haoneng Luo and Xian Shi and Mengzhe Chen and Yabin Li and Lingyun Zuo and Zhihao Du and Zhangyu Xiao and Shiliang Zhang},
|
||||
title={FunASR: A Fundamental End-to-End Speech Recognition Toolkit},
|
||||
year={2023},
|
||||
booktitle={INTERSPEECH},
|
||||
author = {Zhifu Gao and Zerui Li and Jiaming Wang and Haoneng Luo and Xian Shi and Mengzhe Chen and Yabin Li and Lingyun Zuo and Zhihao Du and Zhangyu Xiao and Shiliang Zhang},
|
||||
title = {FunASR: A Fundamental End-to-End Speech Recognition Toolkit},
|
||||
booktitle = {INTERSPEECH},
|
||||
year = {2023},
|
||||
}
|
||||
```
|
||||
|
||||
## 许可证
|
||||
|
||||
代码：[MIT](./LICENSE) · 模型权重：[FunASR 模型许可](./MODEL_LICENSE)
|
||||
代码：[MIT 许可证](./LICENSE) · 模型权重：[FunASR 模型许可](./MODEL_LICENSE)（允许商业使用，需注明出处）
|
||||
|
||||
Reference in New Issue
Block a user