Skip to content

Open-source speech recognition toolkit for training, inference, streaming ASR, VAD, punctuation, speaker diarization pipelines, and OpenAI-compatible/MCP serving.

docs: distinguish Fun-ASR checkpoints in README

游雁f7b20041e23f
diff --git a/README.md b/README.md
index 7c8c0c9bb..1ded43143 100644
--- a/README.md
+++ b/README.md
@@ -5,8 +5,8 @@
</p>
<p align="center">
- <strong>Industrial speech recognition. Up to 340x realtime, 26x faster than Whisper. 50+ languages.</strong><br>
- <em>Speaker diarization · Emotion detection · Streaming · One API call</em>
+ <strong>Industrial speech recognition toolkit for offline, streaming, and edge deployment.</strong><br>
+ <em>ASR · VAD · punctuation · speaker pipelines · emotion and audio-event models · OpenAI-compatible serving</em>
</p>
<p align="center">
@@ -37,7 +37,7 @@ pip install torch torchaudio
pip install funasr
```
-**Flagship model — Fun-ASR-Nano** (LLM-ASR, 31 languages; the default recommendation, needs a GPU):
+**Flagship model — Fun-ASR-Nano** (LLM-ASR for Chinese, English, and Japanese, including Chinese dialects; needs a GPU):
```python
from funasr import AutoModel
@@ -48,7 +48,14 @@ print(result[0]["text"])
# 欢迎大家来体验达摩院推出的语音识别模型。
```
-On CPU (or for multilingual + emotion in one pass), use **SenseVoice** — which also returns speaker diarization and timestamps:
+For the separate 31-language checkpoint, use
+[Fun-ASR-MLT-Nano-2512](https://huggingface.co/FunAudioLLM/Fun-ASR-MLT-Nano-2512).
+Language coverage is checkpoint-specific.
+
+On CPU (or for five-language ASR plus emotion and audio-event tags), use
+**SenseVoiceSmall**. The pipeline below composes SenseVoiceSmall with FSMN-VAD
+and CAM++; diarization is provided by the separate CAM++ model, not by the
+SenseVoiceSmall checkpoint:
```python
from funasr import AutoModel
@@ -60,7 +67,7 @@ result = model.generate(
batch_size_s=300,
)
-# One call returns VAD segments with speaker id + timestamps — render them however you like:
+# The AutoModel pipeline returns VAD segments with speaker ids and timestamps:
for seg in result[0]["sentence_info"]:
print(f"[{seg['start']/1000:.1f}s] Speaker {seg['spk']}: {rich_transcription_postprocess(seg['sentence'])}")
```
@@ -70,7 +77,8 @@ for seg in result[0]["sentence_info"]:
[0.6s] Speaker 0: 欢迎大家来体验达摩院推出的语音识别模型
```
-That's it. **One model, one call** — VAD segmentation, speech recognition, punctuation, speaker diarization all happen automatically.
+One `AutoModel` pipeline call coordinates the configured ASR, VAD, and speaker
+models and returns the combined result.
### Scale & deploy the flagship
@@ -89,14 +97,19 @@ results = model.generate(["audio1.wav", "audio2.wav"], language="auto")
### Why FunASR?
-Whisper is a single model; **FunASR is a toolkit** — you pick the right model per job: **Fun-ASR-Nano** (flagship LLM-ASR, GPU, 340x realtime with vLLM, 31 languages), **SenseVoice** (CPU-friendly, + emotion & audio events), **Paraformer** (low-latency streaming). The table shows what the toolkit delivers vs one Whisper model — each capability is labelled with the model that provides it:
+Whisper is a single model; **FunASR is a toolkit** — you pick the right model
+per job: **Fun-ASR-Nano** (Chinese, English, Japanese, and Chinese dialects;
+GPU), **Fun-ASR-MLT-Nano** (31 languages), **SenseVoiceSmall** (five-language
+ASR plus emotion and audio events), and **Paraformer** (low-latency streaming).
+The table shows toolkit-level capabilities and names the model or pipeline that
+provides each one:
| | FunASR (toolkit) | Whisper | Cloud APIs |
|---|---|---|---|
| Top speed | **340x realtime** (Fun-ASR-Nano + vLLM) | 13x realtime | ~1x realtime |
-| Speaker ID | ✅ Built-in | ❌ Needs pyannote | ✅ Extra cost |
+| Speaker ID | ✅ via VAD + CAM++ pipeline | ❌ Needs pyannote | ✅ Extra cost |
| Emotion | ✅ via SenseVoice | ❌ | ❌ |
-| Languages | 50+ (Qwen3-ASR 52, Nano 31) | 57 | Varies |
+| Languages | Checkpoint-specific (for example Qwen3-ASR 52, MLT-Nano 31, Nano zh/en/ja) | 57 | Varies |
| Streaming | ✅ WebSocket (Paraformer) | ❌ | ✅ |
| CPU viable | ✅ 17x realtime (SenseVoice) | ❌ Too slow | N/A |
| Self-hosted | ✅ Yes (toolkit: MIT; model licenses vary) | ✅ MIT license | ❌ Cloud only |
@@ -128,7 +141,8 @@ Requirements: Python ≥ 3.8. Install PyTorch + torchaudio first ([pytorch.org](
| Model | Task | Languages | Params | Links |
|-------|------|-----------|--------|-------|
-| **Fun-ASR-Nano** | ASR + timestamps | 31 languages | 800M | [⭐](https://www.modelscope.cn/models/FunAudioLLM/Fun-ASR-Nano-2512) [🤗](https://huggingface.co/FunAudioLLM/Fun-ASR-Nano-2512) |
+| **Fun-ASR-Nano** | ASR | zh/en/ja + Chinese dialects | 800M | [⭐](https://www.modelscope.cn/models/FunAudioLLM/Fun-ASR-Nano-2512) [🤗](https://huggingface.co/FunAudioLLM/Fun-ASR-Nano-2512) |
+| **Fun-ASR-MLT-Nano** | ASR | 31 languages | 800M | [⭐](https://www.modelscope.cn/models/FunAudioLLM/Fun-ASR-MLT-Nano-2512) [🤗](https://huggingface.co/FunAudioLLM/Fun-ASR-MLT-Nano-2512) |
| **SenseVoiceSmall** | ASR + emotion + events | zh/en/ja/ko/yue | 234M | [⭐](https://www.modelscope.cn/models/iic/SenseVoiceSmall) [🤗](https://huggingface.co/FunAudioLLM/SenseVoiceSmall) |
| **Paraformer-zh** | ASR + timestamps | zh/en | 220M | [⭐](https://www.modelscope.cn/models/iic/speech_paraformer-large-vad-punc_asr_nat-zh-cn-16k-common-vocab8404-pytorch/summary) [🤗](https://huggingface.co/funasr/paraformer-zh) |
| Paraformer-zh-streaming | Streaming ASR | zh/en | 220M | [⭐](https://modelscope.cn/models/iic/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-online/summary) [🤗](https://huggingface.co/funasr/paraformer-zh-streaming) |
@@ -269,8 +283,8 @@ llama-funasr-sensevoice -m ./gguf/SenseVoiceSmall-f16.gguf --vad ./gguf/fsmn-vad
- 2026/05/24: **v1.3.3** — `funasr-server` CLI, OpenAI-compatible API, MCP Server for AI agents. `pip install --upgrade funasr`
- 2026/05/20: Added Qwen3-ASR (0.6B/1.7B) — 52 languages, auto detection. [usage](examples/industrial_data_pretraining/qwen3_asr)
- 2026/05/20: Added GLM-ASR-Nano (1.5B) — 17 languages, dialect support. [usage](examples/industrial_data_pretraining/glm_asr)
-- 2026/05/19: Fun-ASR-Nano and SenseVoice now support speaker diarization.
-- 2025/12/15: [Fun-ASR-Nano-2512](https://github.com/FunAudioLLM/Fun-ASR) — 31 languages, tens of millions of hours training.
+- 2026/05/19: Fun-ASR-Nano and SenseVoice can be combined with VAD and CAM++ for speaker diarization.
+- 2025/12/15: [Fun-ASR-Nano-2512](https://github.com/FunAudioLLM/Fun-ASR) — Chinese, English, Japanese, and Chinese dialect support; trained on tens of millions of hours.
<details><summary>Older</summary>
diff --git a/README_ja.md b/README_ja.md
index 31b732c93..d350171b0 100644
--- a/README_ja.md
+++ b/README_ja.md
@@ -5,8 +5,8 @@
</p>
<p align="center">
- <strong>産業グレードの音声認識。最大340倍リアルタイム、Whisperより26倍高速。50以上の言語に対応。</strong><br>
- <em>話者分離 · 感情認識 · ストリーミング · ワンコールで完結</em>
+ <strong>オフライン、ストリーミング、エッジ展開に対応する産業グレードの音声認識ツールキット。</strong><br>
+ <em>ASR · VAD · 句読点 · 話者パイプライン · 感情/音声イベントモデル · OpenAI互換配信</em>
</p>
<p align="center">
@@ -48,7 +48,9 @@ result = model.generate(input="meeting.wav")
[00:07.5 → 00:12.3] 話者0: どうぞ。あと30分あります。
```
-1つのモデル、1回の呼び出し — VADセグメンテーション、音声認識、句読点復元、話者分離がすべて自動で実行されます。
+これは1回の `AutoModel` パイプライン呼び出しですが、SenseVoiceSmall、
+FSMN-VAD、CAM++という独立したモデルを組み合わせています。話者分離は
+SenseVoiceSmall自体ではなく、CAM++によって提供されます。
初めて使う場合は [Colab クイックスタート](./examples/colab/README_ja.md) から試せます。どのモデルを選ぶか迷う場合は [モデル選択ガイド](./docs/model_selection_ja.md) を参照してください。
@@ -58,14 +60,18 @@ result = model.generate(input="meeting.wav")
### なぜFunASRを選ぶのか?
-Whisper は単一モデルですが、**FunASR はツールキット**です——用途に応じてモデルを選べます:**Fun-ASR-Nano**(フラッグシップ LLM-ASR、GPU が必要、vLLM で 340 倍リアルタイム、31 言語)、**SenseVoice**(CPU に優しく、感情・音声イベントも)、**Paraformer**(低遅延ストリーミング)。下の表は、単一の Whisper モデルに対してツールキットが提供できるものです——各機能にはそれを提供するモデルを併記しています:
+Whisper は単一モデルですが、**FunASR はツールキット**です——用途に応じて
+**Fun-ASR-Nano**(中・英・日と中国語方言、GPU)、
+**Fun-ASR-MLT-Nano**(31言語)、**SenseVoiceSmall**(5言語ASRと感情・
+音声イベント)、**Paraformer**(低遅延ストリーミング)を選べます。
+下の表はツールキット全体の機能と、それを提供するモデルまたはパイプラインを示します:
| | FunASR(ツールキット) | Whisper | クラウドAPI |
|---|---|---|---|
| 最高速度 | **340倍リアルタイム**(Fun-ASR-Nano + vLLM) | 13倍リアルタイム | 〜1倍リアルタイム |
-| 話者認識 | ✅ 内蔵 | ❌ pyannoteが必要 | ✅ 追加料金 |
+| 話者認識 | ✅ VAD + CAM++パイプライン | ❌ pyannoteが必要 | ✅ 追加料金 |
| 感情認識 | ✅ SenseVoice による | ❌ | ❌ |
-| 言語数 | 50以上(Qwen3-ASR 52、Nano 31) | 57 | サービスにより異なる |
+| 言語数 | チェックポイントごとに異なる(例:Qwen3-ASR 52、MLT-Nano 31、Nano 中/英/日) | 57 | サービスにより異なる |
| ストリーミング | ✅ WebSocket(Paraformer) | ❌ | ✅ |
| CPU対応 | ✅ 17倍リアルタイム(SenseVoice) | ❌ 遅すぎる | 該当なし |
| セルフホスト | ✅ 対応(ツールキット: MIT、モデルごとに異なる) | ✅ MITライセンス | ❌ クラウドのみ |
@@ -96,7 +102,7 @@ Whisper は単一モデルですが、**FunASR はツールキット**です—
- 2026/05/24:**v1.3.3** — `funasr-server` CLI、OpenAI互換API、MCPサーバー。`pip install --upgrade funasr`
- 2026/05/20:Qwen3-ASR (0.6B/1.7B) 追加 — 52言語対応。
- 2026/05/20:GLM-ASR-Nano (1.5B) 追加 — 17言語、方言対応。
-- 2025/12/15:[Fun-ASR-Nano-2512](https://github.com/FunAudioLLM/Fun-ASR) — 31言語対応。
+- 2025/12/15:[Fun-ASR-Nano-2512](https://github.com/FunAudioLLM/Fun-ASR) — 中・英・日と中国語方言に対応。
---
@@ -116,7 +122,8 @@ pip install funasr
| モデル | タスク | 言語 | パラメータ | リンク |
|--------|--------|------|-----------|--------|
-| **Fun-ASR-Nano** | 認識 + タイムスタンプ | 31言語 | 800M | [⭐](https://www.modelscope.cn/models/FunAudioLLM/Fun-ASR-Nano-2512) [🤗](https://huggingface.co/FunAudioLLM/Fun-ASR-Nano-2512) |
+| **Fun-ASR-Nano** | 認識 | 中/英/日 + 中国語方言 | 800M | [⭐](https://www.modelscope.cn/models/FunAudioLLM/Fun-ASR-Nano-2512) [🤗](https://huggingface.co/FunAudioLLM/Fun-ASR-Nano-2512) |
+| **Fun-ASR-MLT-Nano** | 認識 | 31言語 | 800M | [⭐](https://www.modelscope.cn/models/FunAudioLLM/Fun-ASR-MLT-Nano-2512) [🤗](https://huggingface.co/FunAudioLLM/Fun-ASR-MLT-Nano-2512) |
| **SenseVoiceSmall** | 認識 + 感情 + イベント | 中/英/日/韓/粤 | 234M | [⭐](https://www.modelscope.cn/models/iic/SenseVoiceSmall) [🤗](https://huggingface.co/FunAudioLLM/SenseVoiceSmall) |
| **Paraformer-zh** | 認識 + タイムスタンプ | 中/英 | 220M | [⭐](https://www.modelscope.cn/models/damo/speech_paraformer-large-vad-punc_asr_nat-zh-cn-16k-common-vocab8404-pytorch/summary) [🤗](https://huggingface.co/funasr/paraformer-zh) |
| Qwen3-ASR | 認識、52言語 | 多言語 | 1.7B | [使用法](examples/industrial_data_pretraining/qwen3_asr) |
diff --git a/README_ko.md b/README_ko.md
index 72c47a889..cfcc39148 100644
--- a/README_ko.md
+++ b/README_ko.md
@@ -5,8 +5,8 @@
</p>
<p align="center">
- <strong>산업용 음성인식. 최대 340배 실시간, Whisper보다 26배 빠름. 50개 이상 언어 지원.</strong><br>
- <em>화자 분리 · 감정 인식 · 스트리밍 · 한 번의 호출로 해결</em>
+ <strong>오프라인, 스트리밍 및 엣지 배포를 위한 산업용 음성 인식 툴킷.</strong><br>
+ <em>ASR · VAD · 구두점 · 화자 파이프라인 · 감정 및 오디오 이벤트 모델 · OpenAI 호환 서빙</em>
</p>
<p align="center">
@@ -48,7 +48,9 @@ result = model.generate(input="meeting.wav")
[00:07.5 → 00:12.3] 화자0: 말씀하세요. 30분 남았습니다.
```
-하나의 모델, 한 번의 호출 — VAD 분할, 음성인식, 구두점 복원, 화자 분리가 모두 자동으로 수행됩니다.
+한 번의 `AutoModel` 파이프라인 호출이지만, SenseVoiceSmall, FSMN-VAD,
+CAM++라는 독립된 모델을 조합합니다. 화자 분리는 SenseVoiceSmall 자체가 아니라
+CAM++에서 제공합니다.
처음 사용한다면 [Colab 빠른 시작](./examples/colab/README_ko.md)으로 먼저 확인할 수 있습니다. 어떤 모델을 선택할지 고민된다면 [모델 선택 가이드](./docs/model_selection_ko.md)를 참고하세요.
@@ -58,14 +60,18 @@ result = model.generate(input="meeting.wav")
### 왜 FunASR인가?
-Whisper는 단일 모델이지만, **FunASR는 툴킷**입니다 — 용도에 맞는 모델을 고르세요: **Fun-ASR-Nano**(플래그십 LLM-ASR, GPU 필요, vLLM로 340배 실시간, 31개 언어), **SenseVoice**(CPU 친화적, 감정·오디오 이벤트 포함), **Paraformer**(저지연 스트리밍). 아래 표는 단일 Whisper 모델 대비 툴킷이 제공하는 것이며, 각 기능에는 이를 제공하는 모델을 표기했습니다:
+Whisper는 단일 모델이지만, **FunASR는 툴킷**입니다. 용도에 맞게
+**Fun-ASR-Nano**(중국어/영어/일본어 및 중국어 방언, GPU),
+**Fun-ASR-MLT-Nano**(31개 언어), **SenseVoiceSmall**(5개 언어 ASR와
+감정·오디오 이벤트), **Paraformer**(저지연 스트리밍)를 선택하세요.
+아래 표는 툴킷 전체의 기능과 이를 제공하는 모델 또는 파이프라인을 보여 줍니다:
| | FunASR(툴킷) | Whisper | 클라우드 API |
|---|---|---|---|
| 최고 속도 | **340배 실시간**(Fun-ASR-Nano + vLLM) | 13배 실시간 | ~1배 실시간 |
-| 화자 인식 | ✅ 내장 | ❌ pyannote 필요 | ✅ 추가 비용 |
+| 화자 인식 | ✅ VAD + CAM++ 파이프라인 | ❌ pyannote 필요 | ✅ 추가 비용 |
| 감정 인식 | ✅ SenseVoice 제공 | ❌ | ❌ |
-| 언어 수 | 50개 이상(Qwen3-ASR 52, Nano 31) | 57개 | 서비스마다 다름 |
+| 언어 수 | 체크포인트별 상이(예: Qwen3-ASR 52, MLT-Nano 31, Nano 중/영/일) | 57개 | 서비스마다 다름 |
| 스트리밍 | ✅ WebSocket(Paraformer) | ❌ | ✅ |
| CPU 사용 | ✅ 17배 실시간(SenseVoice) | ❌ 너무 느림 | 해당 없음 |
| 자체 호스팅 | ✅ 지원 (툴킷: MIT, 모델별 상이) | ✅ MIT 라이선스 | ❌ 클라우드만 |
@@ -96,7 +102,7 @@ Whisper는 단일 모델이지만, **FunASR는 툴킷**입니다 — 용도에
- 2026/05/24: **v1.3.3** — `funasr-server` CLI, OpenAI 호환 API, MCP 서버. `pip install --upgrade funasr`
- 2026/05/20: Qwen3-ASR (0.6B/1.7B) 추가 — 52개 언어 지원.
- 2026/05/20: GLM-ASR-Nano (1.5B) 추가 — 17개 언어, 방언 지원.
-- 2025/12/15: [Fun-ASR-Nano-2512](https://github.com/FunAudioLLM/Fun-ASR) — 31개 언어 지원.
+- 2025/12/15: [Fun-ASR-Nano-2512](https://github.com/FunAudioLLM/Fun-ASR) — 중국어, 영어, 일본어 및 중국어 방언 지원.
---
@@ -116,7 +122,8 @@ pip install funasr
| 모델 | 작업 | 언어 | 파라미터 | 링크 |
|------|------|------|---------|------|
-| **Fun-ASR-Nano** | 인식 + 타임스탬프 | 31개 언어 | 800M | [⭐](https://www.modelscope.cn/models/FunAudioLLM/Fun-ASR-Nano-2512) [🤗](https://huggingface.co/FunAudioLLM/Fun-ASR-Nano-2512) |
+| **Fun-ASR-Nano** | 인식 | 중/영/일 + 중국어 방언 | 800M | [⭐](https://www.modelscope.cn/models/FunAudioLLM/Fun-ASR-Nano-2512) [🤗](https://huggingface.co/FunAudioLLM/Fun-ASR-Nano-2512) |
+| **Fun-ASR-MLT-Nano** | 인식 | 31개 언어 | 800M | [⭐](https://www.modelscope.cn/models/FunAudioLLM/Fun-ASR-MLT-Nano-2512) [🤗](https://huggingface.co/FunAudioLLM/Fun-ASR-MLT-Nano-2512) |
| **SenseVoiceSmall** | 인식 + 감정 + 이벤트 | 중/영/일/한/광둥어 | 234M | [⭐](https://www.modelscope.cn/models/iic/SenseVoiceSmall) [🤗](https://huggingface.co/FunAudioLLM/SenseVoiceSmall) |
| **Paraformer-zh** | 인식 + 타임스탬프 | 중/영 | 220M | [⭐](https://www.modelscope.cn/models/damo/speech_paraformer-large-vad-punc_asr_nat-zh-cn-16k-common-vocab8404-pytorch/summary) [🤗](https://huggingface.co/funasr/paraformer-zh) |
| Qwen3-ASR | 인식, 52개 언어 | 다국어 | 1.7B | [사용법](examples/industrial_data_pretraining/qwen3_asr) |
diff --git a/README_zh.md b/README_zh.md
index 2f1f07bbc..274ca8995 100644
--- a/README_zh.md
+++ b/README_zh.md
@@ -5,8 +5,8 @@
</p>
<p align="center">
- <strong>工业级语音识别。最高 340 倍实时,比 Whisper 快 26 倍。支持 50+ 语言。</strong><br>
- <em>说话人分离 · 情感识别 · 流式转写 · 一次调用搞定</em>
+ <strong>面向离线、流式与边缘部署的工业级语音识别工具箱。</strong><br>
+ <em>ASR · VAD · 标点 · 说话人 pipeline · 情感与音频事件模型 · OpenAI 兼容服务</em>
</p>
<p align="center">
@@ -43,7 +43,7 @@ from funasr.utils.postprocess_utils import rich_transcription_postprocess
model = AutoModel(model="iic/SenseVoiceSmall", vad_model="fsmn-vad", spk_model="cam++", device="cuda")
result = model.generate(input="https://isv-data.oss-cn-hangzhou.aliyuncs.com/ics/MaaS/ASR/test_audio/asr_example_zh.wav")
-# 一次调用即返回带说话人 id 和时间戳的 VAD 分段,可自由渲染:
+# AutoModel pipeline 返回带说话人 id 和时间戳的 VAD 分段:
for seg in result[0]["sentence_info"]:
print(f"[{seg['start']/1000:.1f}s] 说话人{seg['spk']}: {rich_transcription_postprocess(seg['sentence'])}")
```
@@ -53,11 +53,14 @@ for seg in result[0]["sentence_info"]:
[0.6s] 说话人0: 欢迎大家来体验达摩院推出的语音识别模型
```
-一个模型、一次调用 — VAD 分段、语音识别、标点恢复、说话人分离全部自动完成。
+这是一次 `AutoModel` pipeline 调用,实际组合了 SenseVoiceSmall、FSMN-VAD
+和 CAM++ 三个独立模型;说话人分离由 CAM++ 提供,并非 SenseVoiceSmall
+checkpoint 的内置输出。
### LLM 语音识别:Fun-ASR-Nano
-追求更高精度、支持 31 种语言(含中文方言),使用 [Fun-ASR-Nano](https://github.com/FunAudioLLM/Fun-ASR) — SenseVoice 编码器 + Qwen3-0.6B 解码器的 LLM-based ASR:
+Fun-ASR-Nano 是基于 SenseVoice 编码器 + Qwen3-0.6B 解码器的 LLM-ASR,
+支持中文、英语、日语,以及 7 种中文方言和 26 种地域口音:
```python
from funasr import AutoModel
@@ -66,7 +69,11 @@ model = AutoModel(model="FunAudioLLM/Fun-ASR-Nano-2512", vad_model="fsmn-vad", d
result = model.generate(input="https://isv-data.oss-cn-hangzhou.aliyuncs.com/ics/MaaS/ASR/test_audio/asr_example_zh.wav")
```
-使用 vLLM 加速(批量处理快 16 倍):
+需要 31 语种时,请改用独立的
+[Fun-ASR-MLT-Nano-2512](https://huggingface.co/FunAudioLLM/Fun-ASR-MLT-Nano-2512)
+checkpoint。不同 checkpoint 的语言范围不能混用。
+
+使用 vLLM 做高吞吐批处理:
```python
from funasr.auto.auto_model_vllm import AutoModelVLLM
@@ -81,14 +88,18 @@ results = model.generate(["audio1.wav", "audio2.wav"], language="auto")
### 为什么选 FunASR?
-Whisper 是单个模型,**FunASR 是一个工具箱**——按场景挑模型:**Fun-ASR-Nano**(旗舰 LLM-ASR,需 GPU,vLLM 下 340 倍实时,31 种语言)、**SenseVoice**(CPU 友好,额外带情感与音频事件)、**Paraformer**(低延迟流式)。下表是工具箱相比单个 Whisper 模型能提供什么——每项能力都标注了由哪个模型提供:
+Whisper 是单个模型,**FunASR 是一个工具箱**——按场景挑模型:
+**Fun-ASR-Nano**(中/英/日及中文方言,需 GPU)、
+**Fun-ASR-MLT-Nano**(31 语种)、**SenseVoiceSmall**(五语种 ASR,
+并返回情感与音频事件标签)、**Paraformer**(低延迟流式)。下表展示的是
+工具箱级能力,并标明由哪个模型或 pipeline 提供:
| | FunASR(工具箱) | Whisper | 云端 API |
|---|---|---|---|
| 最高速度 | **340 倍实时**(Fun-ASR-Nano + vLLM) | 13 倍实时 | ~1 倍实时 |
-| 说话人识别 | ✅ 内置 | ❌ 需要 pyannote | ✅ 额外付费 |
+| 说话人识别 | ✅ 由 VAD + CAM++ pipeline 提供 | ❌ 需要 pyannote | ✅ 额外付费 |
| 情感识别 | ✅ 由 SenseVoice 提供 | ❌ | ❌ |
-| 语言数 | 50+(Qwen3-ASR 52、Nano 31) | 57 | 因服务而异 |
+| 语言数 | 取决于 checkpoint(例如 Qwen3-ASR 52、MLT-Nano 31、Nano 中/英/日) | 57 | 因服务而异 |
| 流式识别 | ✅ WebSocket(Paraformer) | ❌ | ✅ |
| CPU 可用 | ✅ 17 倍实时(SenseVoice) | ❌ 太慢 | 不适用 |
| 私有部署 | ✅ 支持(工具箱 MIT;模型协议各异) | ✅ MIT 开源 | ❌ 仅云端 |
@@ -123,8 +134,8 @@ Whisper 是单个模型,**FunASR 是一个工具箱**——按场景挑模型
- 2026/05/24:**v1.3.3** — `funasr-server` 命令行工具、OpenAI 兼容 API、MCP 服务。`pip install --upgrade funasr`
- 2026/05/20:新增 Qwen3-ASR (0.6B/1.7B),52 种语言自动检测。[使用方法](examples/industrial_data_pretraining/qwen3_asr)
- 2026/05/20:新增 GLM-ASR-Nano (1.5B),17 种语言,方言优化。[使用方法](examples/industrial_data_pretraining/glm_asr)
-- 2026/05/19:Fun-ASR-Nano 和 SenseVoice 支持说话人分离。
-- 2025/12/15:[Fun-ASR-Nano-2512](https://github.com/FunAudioLLM/Fun-ASR) 上线。
+- 2026/05/19:Fun-ASR-Nano 和 SenseVoice 可与 VAD、CAM++ 组合为说话人分离 pipeline。
+- 2025/12/15:[Fun-ASR-Nano-2512](https://github.com/FunAudioLLM/Fun-ASR) 上线,支持中/英/日及中文方言。
<details><summary>更早</summary>
@@ -160,7 +171,8 @@ pip install -e ./
| 模型 | 任务 | 语言 | 参数量 | 链接 |
|------|------|------|--------|------|
-| **Fun-ASR-Nano** | 识别 + 时间戳 | 31 种语言 | 800M | [⭐](https://www.modelscope.cn/models/FunAudioLLM/Fun-ASR-Nano-2512) [🤗](https://huggingface.co/FunAudioLLM/Fun-ASR-Nano-2512) |
+| **Fun-ASR-Nano** | 识别 | 中/英/日 + 中文方言 | 800M | [⭐](https://www.modelscope.cn/models/FunAudioLLM/Fun-ASR-Nano-2512) [🤗](https://huggingface.co/FunAudioLLM/Fun-ASR-Nano-2512) |
+| **Fun-ASR-MLT-Nano** | 识别 | 31 种语言 | 800M | [⭐](https://www.modelscope.cn/models/FunAudioLLM/Fun-ASR-MLT-Nano-2512) [🤗](https://huggingface.co/FunAudioLLM/Fun-ASR-MLT-Nano-2512) |
| **SenseVoiceSmall** | 识别 + 情感 + 事件 | 中/英/日/韩/粤 | 234M | [⭐](https://www.modelscope.cn/models/iic/SenseVoiceSmall) [🤗](https://huggingface.co/FunAudioLLM/SenseVoiceSmall) |
| **Paraformer-zh** | 识别 + 时间戳 | 中/英 | 220M | [⭐](https://www.modelscope.cn/models/damo/speech_paraformer-large-vad-punc_asr_nat-zh-cn-16k-common-vocab8404-pytorch/summary) [🤗](https://huggingface.co/funasr/paraformer-zh) |
| Paraformer-zh-streaming | 流式识别 | 中/英 | 220M | [⭐](https://modelscope.cn/models/damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-online/summary) [🤗](https://huggingface.co/funasr/paraformer-zh-streaming) |
@@ -186,7 +198,7 @@ from funasr import AutoModel
model = AutoModel(model="paraformer-zh", vad_model="fsmn-vad", punc_model="ct-punc", spk_model="cam++", device="cuda")
result = model.generate(input="https://isv-data.oss-cn-hangzhou.aliyuncs.com/ics/MaaS/ASR/test_audio/asr_example_zh.wav", hotword="关键词 20")
-# 31 种语言 + 时间戳
+# 中/英/日 + 中文方言
model = AutoModel(model="FunAudioLLM/Fun-ASR-Nano-2512", hub="hf", trust_remote_code=True,
vad_model="fsmn-vad", vad_kwargs={"max_single_segment_time": 30000}, device="cuda")
result = model.generate(input="audio.wav", batch_size=1)