Open-source speech recognition toolkit for training, inference, streaming ASR, VAD, punctuation, speaker diarization pipelines, and OpenAI-compatible/MCP serving.
Python asraudiochineseemotion-recognitionfunasrmcp-servermultilingual-asropenai-compatible-apiparaformerpunctuationpytorchreal-time-asrspeaker-diarizationspeech-recognitionspeech-to-textstreaming-asrtranscriptionvllmvoice-activity-detectionwhisper-alternative
<!DOCTYPE html>
<html lang="zh">
<head><meta http-equiv="Cache-Control" content="no-cache, no-store, must-revalidate"><meta http-equiv="Pragma" content="no-cache"><meta http-equiv="Expires" content="0">
<meta charset="utf-8">
<meta name="viewport" content="width=device-width,initial-scale=1">
<title>FunASR 快速上手指南 | 语音识别 Python 教程 - 三分钟开始</title>
<meta name="description" content="FunASR 快速上手指南:三分钟学会语音转文字。涵盖离线文件转写、实时流式识别、OpenAI 兼容 API 服务部署,附完整可运行代码。自动语音识别 Python 教程。">
<meta name="keywords" content="自动语音识别, 语音转文字, ASR API, speech recognition python, FunASR 教程, 语音识别 Python, 实时语音识别, OpenAI Whisper 替代">
<link rel="icon" href="logo.png">
<meta property="og:title" content="FunASR 快速上手指南 — 三分钟开始语音识别">
<meta property="og:description" content="离线文件转写、实时流式识别、OpenAI 兼容 API 部署——完整可运行代码,复制即用。">
<meta property="og:type" content="article">
<meta property="og:url" content="https://www.funasr.com/quickstart.html">
<meta property="og:image" content="https://www.funasr.com/logo.png">
<link rel="canonical" href="https://www.funasr.com/quickstart.html">
<link rel="alternate" hreflang="en" href="https://www.funasr.com/en/quickstart.html">
<link rel="alternate" hreflang="zh" href="https://www.funasr.com/quickstart.html">
<link href="https://fonts.googleapis.com/css2?family=Inter:wght@400;500;600;700;800&display=swap" rel="stylesheet">
<script type="application/ld+json">
{
"@context": "https://schema.org",
"@type": "HowTo",
"name": "FunASR 快速上手指南",
"description": "三分钟学会使用 FunASR 进行语音识别:离线文件转写、实时流式识别、OpenAI 兼容 API 服务部署。",
"image": "https://www.funasr.com/logo.png",
"totalTime": "PT3M",
"tool": [
{"@type": "HowToTool", "name": "Python 3.8+"},
{"@type": "HowToTool", "name": "pip"},
{"@type": "HowToTool", "name": "GPU (8GB+ VRAM, 可选)"}
],
"step": [
{
"@type": "HowToStep",
"position": 1,
"name": "安装 FunASR",
"text": "使用 pip 安装 FunASR 及依赖:pip install funasr",
"url": "https://www.funasr.com/quickstart.html#install"
},
{
"@type": "HowToStep",
"position": 2,
"name": "离线文件转写",
"text": "使用 AutoModel 加载旗舰 Fun-ASR-Nano 模型,调用 generate 转写中英日及中文方言/口音,获得 LLM 级别输出。",
"url": "https://www.funasr.com/quickstart.html#file-transcription"
},
{
"@type": "HowToStep",
"position": 3,
"name": "实时流式识别",
"text": "使用 Paraformer 流式模型进行实时语音识别,支持麦克风输入和 WebSocket 协议。",
"url": "https://www.funasr.com/quickstart.html#streaming"
},
{
"@type": "HowToStep",
"position": 4,
"name": "部署 OpenAI 兼容 API",
"text": "一行命令启动 OpenAI 兼容的语音识别 API 服务,支持 curl、Python SDK 和 AI 框架直接调用。",
"url": "https://www.funasr.com/quickstart.html#api-server"
}
]
}
</script>
<style>
:root{--primary:#2563eb;--primary-dark:#1d4ed8;--bg:#fff;--surface:#f8fafc;--border:#e2e8f0;--text:#0f172a;--text-soft:#475569;--text-muted:#94a3b8;--radius:12px}
*{margin:0;padding:0;box-sizing:border-box}
html{scroll-behavior:smooth}
body{font-family:'Inter',-apple-system,BlinkMacSystemFont,sans-serif;color:var(--text);line-height:1.6;-webkit-font-smoothing:antialiased}
a{color:var(--primary);text-decoration:none}
a:hover{text-decoration:underline}
.container{max-width:1120px;margin:0 auto;padding:0 24px}
.nav{position:fixed;top:0;left:0;right:0;z-index:100;background:rgba(255,255,255,0.85);backdrop-filter:blur(12px);border-bottom:1px solid var(--border);padding:14px 0}
.nav .container{display:flex;align-items:center;gap:32px}
.nav-logo{font-size:1.3rem;font-weight:800;color:var(--text);text-decoration:none}
.nav-logo span{color:var(--primary)}
.nav-links{display:flex;gap:24px;margin-left:auto}
.nav-links a{font-size:0.88rem;font-weight:500;color:var(--text-soft);text-decoration:none;transition:color 0.15s}
.nav-links a:hover{color:var(--primary)}
.nav-links a.active{color:var(--primary)}
.nav-btn{padding:8px 18px;background:var(--primary);color:white;border-radius:8px;font-size:0.85rem;font-weight:600;text-decoration:none}
.nav-btn:hover{background:var(--primary-dark);text-decoration:none}
.page-hero{padding:120px 0 60px;background:linear-gradient(180deg,#0f172a 0%,#1e293b 100%);color:white;text-align:center}
.page-hero h1{font-size:2.6rem;font-weight:800;line-height:1.2}
.page-hero h1 span{background:linear-gradient(135deg,#60a5fa,#a78bfa);-webkit-background-clip:text;-webkit-text-fill-color:transparent}
.page-hero p{margin-top:16px;font-size:1.1rem;color:#94a3b8;max-width:640px;margin-left:auto;margin-right:auto}
.page-hero .prereq{margin-top:24px;display:flex;gap:16px;justify-content:center;flex-wrap:wrap}
.page-hero .prereq span{background:rgba(255,255,255,0.08);border:1px solid rgba(255,255,255,0.12);border-radius:8px;padding:6px 16px;font-size:0.82rem;color:#cbd5e1}
section{padding:80px 0}
section.alt{background:var(--surface)}
section.dark{background:#0f172a;color:white}
.section-header{text-align:center;margin-bottom:48px}
.section-header h2{font-size:2rem;font-weight:800}
.section-header p{margin-top:8px;font-size:1.05rem;color:var(--text-soft);max-width:600px;margin-left:auto;margin-right:auto}
.step-number{display:inline-flex;align-items:center;justify-content:center;width:36px;height:36px;border-radius:50%;background:var(--primary);color:white;font-weight:700;font-size:0.95rem;margin-right:12px;flex-shrink:0}
.guide-section{max-width:820px;margin:0 auto 64px}
.guide-section h2{display:flex;align-items:center;font-size:1.5rem;font-weight:800;margin-bottom:8px}
.guide-section .subtitle{color:var(--text-soft);font-size:0.95rem;margin-bottom:24px;padding-left:48px}
.code-block{background:#1e293b;border-radius:12px;padding:24px;color:#e2e8f0;font-family:'SF Mono',Monaco,Consolas,monospace;font-size:0.82rem;line-height:1.9;overflow-x:auto;margin-bottom:16px;position:relative}
.code-block .c{color:#64748b}
.code-block .p{color:#22c55e}
.code-block .s{color:#86efac}
.code-block .n{color:#fbbf24}
.code-block .k{color:#7dd3fc}
.output-block{background:#f0fdf4;border:1px solid #bbf7d0;border-radius:12px;padding:20px;font-size:0.85rem;color:#166534;line-height:1.7;margin-bottom:20px}
.output-block strong{display:block;font-size:0.75rem;color:#15803d;text-transform:uppercase;letter-spacing:0.5px;margin-bottom:8px}
.tip-block{background:#eff6ff;border:1px solid #bfdbfe;border-radius:12px;padding:20px;font-size:0.85rem;color:#1e40af;line-height:1.7;margin-bottom:20px}
.tip-block strong{display:block;font-size:0.75rem;color:#1d4ed8;text-transform:uppercase;letter-spacing:0.5px;margin-bottom:8px}
.next-grid{display:grid;grid-template-columns:repeat(2,1fr);gap:16px;max-width:720px;margin:0 auto}
.next-card{display:flex;align-items:center;gap:14px;padding:20px 24px;border-radius:var(--radius);border:1px solid var(--border);background:white;text-decoration:none;color:var(--text);transition:all 0.2s}
.next-card:hover{border-color:var(--primary);box-shadow:0 8px 20px rgba(37,99,235,0.06);transform:translateY(-2px);text-decoration:none}
.next-icon{font-size:1.5rem;flex-shrink:0}
.next-card h3{font-size:0.95rem;font-weight:700;margin-bottom:2px}
.next-card p{font-size:0.8rem;color:var(--text-soft)}
footer{background:#0f172a;color:#94a3b8;padding:48px 0 32px}
.footer-grid{display:grid;grid-template-columns:2fr 1fr 1fr 1fr;gap:40px;margin-bottom:32px}
.footer-brand h3{color:white;font-size:1.2rem;margin-bottom:8px}
.footer-brand p{font-size:0.85rem;line-height:1.7}
.footer-col h4{color:white;font-size:0.85rem;margin-bottom:12px;font-weight:600}
.footer-col a{display:block;font-size:0.83rem;color:#94a3b8;margin-bottom:8px;text-decoration:none}
.footer-col a:hover{color:white}
.footer-bottom{border-top:1px solid #1e293b;padding-top:24px;font-size:0.8rem;text-align:center}
@media(max-width:900px){.footer-grid{grid-template-columns:1fr 1fr}.nav-links{display:none}.next-grid{grid-template-columns:1fr}}
@media(max-width:600px){.page-hero h1{font-size:1.8rem}.guide-section h2{font-size:1.2rem}.next-grid{grid-template-columns:1fr}}
</style>
</head>
<body>
<!-- Nav -->
<nav class="nav"><div class="container">
<a href="/" class="nav-logo">Fun<span>ASR</span></a>
<div class="nav-links"><a href="/#capabilities">核心能力</a><a href="/#usage">如何使用</a><a href="/#benchmark">性能</a><a href="/#demo">演示</a><a href="https://modelscope.github.io/FunASR/zh/" target="_blank">文档</a><a href="quickstart.html">快速上手</a><a href="models.html">模型</a><a href="ecosystem.html">生态</a><a href="/blog/">技术博客</a><a href="vs-whisper.html">对比 Whisper</a><a href="/donors.html">功德榜</a></div>
<a href="/en/quickstart.html" style="font-size:0.85rem;font-weight:500;color:var(--text-soft);text-decoration:none">EN</a>
<a href="/go/github" target="_blank" class="nav-btn">GitHub ★</a>
</div></nav>
<!-- Hero -->
<section class="page-hero">
<h1><span>快速上手</span>指南</h1>
<p>三分钟掌握 FunASR 的三种核心用法:文件转写、实时识别、API 服务部署。所有代码可直接复制运行。</p>
<div class="prereq">
<span>Python 3.8+</span>
<span>GPU 8GB+(可选,CPU 也能用)</span>
<span>Linux / macOS / Windows</span>
</div>
</section>
<!-- Install -->
<section id="install"><div class="container">
<div class="guide-section">
<h2><div class="step-number">0</div>安装</h2>
<p class="subtitle">一条命令安装所有依赖。GPU 用户安装 vllm 可获得 16 倍加速。</p>
<div class="code-block">
<span class="c"># 基础安装(CPU / GPU 均可)</span><br>
<span class="p">$</span> pip install funasr<br>
<br>
<span class="c"># 完整安装(含 API 服务 + vLLM 加速,推荐 GPU 用户)</span><br>
<span class="p">$</span> pip install funasr vllm fastapi uvicorn python-multipart
</div>
<div class="tip-block">
<strong>提示</strong>
如果你只需要做一次文件转写,安装 <code>funasr</code> 即可。如果需要部署 API 服务或追求极致速度,建议完整安装。
</div>
</div>
</div></section>
<!-- Section 1: File Transcription -->
<section class="alt" id="file-transcription"><div class="container">
<div class="guide-section">
<h2><div class="step-number">1</div>离线文件转写</h2>
<p class="subtitle">上传音频文件,获得完整转录文本。SenseVoiceSmall 支持中、粤、英、日、韩五种语言,并返回情感和音频事件标签。</p>
<div class="code-block">
<span class="k">from</span> funasr <span class="k">import</span> AutoModel<br>
<br>
<span class="c"># Fun-ASR-Nano:旗舰 LLM-ASR,中/英/日 + 中文方言/口音(需 GPU)</span><br>
model = AutoModel(model=<span class="s">"FunAudioLLM/Fun-ASR-Nano-2512"</span>, device=<span class="s">"cuda"</span>)<br>
result = model.generate(input=<span class="s">"meeting.wav"</span>)<br>
print(result[<span class="n">0</span>][<span class="s">"text"</span>])<br>
<br>
<span class="c"># 需要 31 语种时改用独立 checkpoint:FunAudioLLM/Fun-ASR-MLT-Nano-2512</span><br>
<span class="c"># CPU 场景用 SenseVoice:AutoModel(model="iic/SenseVoiceSmall", vad_model="fsmn-vad")</span>
</div>
<div class="output-block">
<strong>输出结果</strong>
今天我们讨论三个议题。好的,第一个是关于 Q3 计划。请开始吧,我们有半小时。
</div>
<div class="code-block">
<span class="c"># 批量处理多个文件</span><br>
results = model.generate(input=[<span class="s">"file1.wav"</span>, <span class="s">"file2.mp3"</span>, <span class="s">"file3.flac"</span>])<br>
<span class="k">for</span> r <span class="k">in</span> results:<br>
print(r[<span class="s">"text"</span>])
</div>
<div class="tip-block">
<strong>支持的音频格式</strong>
WAV、MP3、FLAC、AAC、OGG、MP4 视频——几乎所有常见音频和视频格式都可以直接处理,无需手动转码。
</div>
</div>
</div></section>
<!-- Section 2: Real-time Streaming -->
<section id="streaming"><div class="container">
<div class="guide-section">
<h2><div class="step-number">2</div>实时流式识别</h2>
<p class="subtitle">麦克风或音频流输入,边说边出结果。适合实时字幕、直播转写、语音助手。</p>
<div class="tip-block">
<strong>模型怎么选</strong>
旗舰 <strong>Fun-ASR-Nano</strong>(LLM-ASR)是<strong>离线</strong>模型,是文件转写和 OpenAI 兼容 API 服务的默认选择(本页第 1、3 节)。而真正<strong>边说边出的低延迟实时流式</strong>,用 FunASR 专门的流式模型 <code>paraformer-zh-streaming</code> —— 即下方示例。同一套工业级 FunASR,按场景搭配模型。
</div>
<div class="code-block">
<span class="k">from</span> funasr <span class="k">import</span> AutoModel<br>
<br>
<span class="c"># Paraformer 流式模型</span><br>
model = AutoModel(model=<span class="s">"paraformer-zh-streaming"</span>, vad_model=<span class="s">"fsmn-vad"</span>)<br>
<br>
<span class="c"># 流式推理:chunk_size = [5, 10, 5] 表示 600ms 前瞻</span><br>
chunk_size = [<span class="n">5</span>, <span class="n">10</span>, <span class="n">5</span>]<br>
cache = {}<br>
<br>
<span class="c"># 模拟流式输入(实际使用时替换为麦克风采集)</span><br>
<span class="k">import</span> soundfile <span class="k">as</span> sf<br>
speech, sr = sf.read(<span class="s">"meeting.wav"</span>)<br>
chunk_stride = chunk_size[<span class="n">1</span>] * <span class="n">960</span> <span class="c"># 每个 chunk 的采样点数</span><br>
<br>
<span class="k">for</span> i <span class="k">in</span> range(<span class="n">0</span>, len(speech), chunk_stride):<br>
chunk = speech[i:i+chunk_stride]<br>
is_final = (i + chunk_stride >= len(speech))<br>
result = model.generate(<br>
input=chunk, cache=cache,<br>
is_final=is_final, chunk_size=chunk_size<br>
)<br>
print(result[<span class="n">0</span>][<span class="s">"text"</span>], end=<span class="s">""</span>, flush=<span class="k">True</span>)
</div>
<div class="output-block">
<strong>实时输出(逐步更新)</strong>
今天我们讨论...三个议题。好的,第一个是...关于 Q3 计划。
</div>
<div class="tip-block">
<strong>WebSocket 实时服务</strong>
如果需要浏览器端实时识别,FunASR 提供完整的 WebSocket 服务方案。参见 <a href="/go/github" target="_blank">GitHub 仓库</a> 中的 <code>examples/</code> 目录获取 WebSocket 服务端和浏览器客户端代码。
</div>
</div>
</div></section>
<!-- Section 3: API Server -->
<section class="alt" id="api-server"><div class="container">
<div class="guide-section">
<h2><div class="step-number">3</div>OpenAI 兼容 API 服务</h2>
<p class="subtitle">一行命令部署 OpenAI 兼容的语音识别 API。LangChain、Dify、Coze 等 AI 框架可零改动对接。</p>
<div class="code-block">
<span class="c"># 安装并启动服务</span><br>
<span class="p">$</span> pip install funasr vllm fastapi uvicorn python-multipart<br>
<span class="p">$</span> funasr-server --device cuda<br>
<br>
<span class="c"># 服务启动后,端口默认为 8899</span>
</div>
<div class="code-block">
<span class="c"># 方式一:curl 调用</span><br>
<span class="p">$</span> curl -X POST http://localhost:<span class="n">8899</span>/v1/audio/transcriptions \<br>
-F <span class="s">"[email protected]"</span> \<br>
-F <span class="s">"model=iic/SenseVoiceSmall"</span>
</div>
<div class="code-block">
<span class="c"># 方式二:OpenAI Python SDK(与 Whisper API 完全相同)</span><br>
<span class="k">from</span> openai <span class="k">import</span> OpenAI<br>
<br>
client = OpenAI(base_url=<span class="s">"http://localhost:8899/v1"</span>, api_key=<span class="s">"x"</span>)<br>
result = client.audio.transcriptions.create(<br>
model=<span class="s">"fun-asr-nano"</span>,<br>
file=open(<span class="s">"audio.wav"</span>, <span class="s">"rb"</span>),<br>
response_format=<span class="s">"verbose_json"</span><br>
)<br>
print(result.text)
</div>
<div class="output-block">
<strong>返回 JSON</strong>
{"text": "今天我们讨论三个议题。", "segments": [{"start": 1.7, "end": 5.5, "text": "今天我们讨论三个议题。"}], "duration": 12.1}
</div>
</div>
</div></section>
<!-- MCP Server -->
<section id="mcp"><div class="container">
<div class="guide-section">
<h2><div class="step-number">+</div>MCP 服务:AI 助手集成</h2>
<p class="subtitle">启用 MCP 协议后,Claude、Cursor、Windsurf 等 AI 助手可直接调用 FunASR 进行语音识别。</p>
<div class="code-block">
<span class="c"># 启动带 MCP 支持的服务</span><br>
<span class="p">$</span> funasr-server --device cuda --enable-mcp<br>
<br>
<span class="c"># 现在 AI 助手可以通过 MCP 协议调用语音识别能力</span><br>
<span class="c"># 适用于 Claude Desktop、Cursor、Windsurf 等工具</span>
</div>
<div class="tip-block">
<strong>什么是 MCP?</strong>
MCP (Model Context Protocol) 是 AI 助手调用外部工具的标准协议。启用后,你的 AI 编程助手可以直接"听懂"音频文件,实现语音驱动的开发工作流。
</div>
</div>
</div></section>
<!-- Next Steps -->
<section class="alt" id="next"><div class="container">
<div class="section-header">
<h2>下一步</h2>
<p>深入了解 FunASR 的更多能力</p>
</div>
<div class="next-grid">
<a class="next-card" href="/go/github" target="_blank">
<div class="next-icon">⭐</div>
<div><h3>GitHub 仓库</h3><p>源代码、示例、Issue 反馈</p></div>
</a>
<a class="next-card" href="models.html">
<div class="next-icon">🧠</div>
<div><h3>模型列表</h3><p>浏览所有可用模型及其能力</p></div>
</a>
<a class="next-card" href="ecosystem.html">
<div class="next-icon">🔗</div>
<div><h3>生态系统</h3><p>FunClip、CosyVoice 等配套工具</p></div>
</a>
<a class="next-card" href="vs-whisper.html">
<div class="next-icon">⚡</div>
<div><h3>对比 Whisper</h3><p>性能、准确率、功能对比</p></div>
</a>
</div>
</div></section>
<!-- Footer -->
<footer><div class="container">
<div class="footer-grid">
<div class="footer-brand">
<h3>FunASR</h3>
<p>通义实验室出品的工业级语音识别服务。语音理解一站式解决方案,为生产环境而生。</p>
</div>
<div class="footer-col"><h4>文档</h4>
<a href="quickstart.html">快速上手</a>
<a href="https://modelscope.github.io/FunASR/zh/training.html">训练微调</a>
<a href="https://github.com/modelscope/FunASR/blob/main/docs/vllm_guide_zh_v2.md">vLLM 指南</a>
<a href="https://modelscope.github.io/FunASR/api.html">API 参考</a>
</div>
<div class="footer-col"><h4>生态</h4>
<a href="/go/fun-asr">Fun-ASR-Nano</a>
<a href="/go/sensevoice">SenseVoice</a>
<a href="/go/funclip">FunClip</a>
<a href="https://github.com/QwenAudio/CosyVoice">CosyVoice</a>
</div>
<div class="footer-col"><h4>社区</h4>
<a href="/go/github">GitHub</a>
<a href="https://github.com/modelscope/FunASR/discussions">Discussions</a>
<a href="https://github.com/modelscope/FunASR/issues">问题反馈</a>
</div>
</div>
<div class="footer-bottom">© 2024-2026 Tongyi Lab, Alibaba Group · <a href="https://github.com/modelscope/FunASR/blob/main/LICENSE" style="color:#94a3b8">MIT License</a></div>
</div></footer>
<script src="/stats/tracker.js" async></script>
</body>
</html>