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
docs(site): publish FunASR 1.4.5 deployment release (#3535)
zhifu gao803ae33eeba1
diff --git a/web-pages/product-site/content/legacy-manifest.json b/web-pages/product-site/content/legacy-manifest.jsonindex 88287c92a..eb55fe8c2 100644--- a/web-pages/product-site/content/legacy-manifest.json+++ b/web-pages/product-site/content/legacy-manifest.json@@ -17,11 +17,12 @@"blog/funasr-v1-3-28-realtime-websocket-subtitles.html": "46a8d4c310e7f76da632d558d057484f48b345d9616873e1416fc68fcb9e9633","blog/funasr-v1-4-0-pypi-release.html": "b5d05ac559435837caf2c3ea2531b581541d283bbecd7c4f3c1d367a2fe81c38","blog/funasr-v1-4-3-pypi-release.html": "6bd73c114653a12e0f8078c4cc7cc89f44cff53f106e12c79b7e6b78b36d922f",+ "blog/funasr-v1-4-5-pypi-llama-cpp-release.html": "14ab5d23f09f61f6bbeb924796e3be9ab53ce87cac5c968322f625c789ae3a8b","blog/funasr-vs-faster-whisper-chinese.html": "bfe9bb8017be80c7e7f4587726f43f6064f1dc4c65e39788f836fdfb0c9789f7","blog/funasr-vs-whisper-benchmark.html": "b7b49adf24d20570abb09b733ce03d4a50a4a0e98e746b4a9320f453e01cce84","blog/funclip-v2-1-0-video-clipping-release.html": "88f6c44e5332d1746c4db0fc97d755ef82f12e46d8f1c1c0ad9351152ff9dfc1","blog/generate-subtitles-srt-vtt-from-audio-video.html": "f1133235673d441cc654310581f59a319f182314ab7c0b0824af87da9f4a0591",- "blog/index.html": "ff7a046ed66e6794e4e0743f531bd4a698efac7519a9414f6d3cd76672eaf33a",+ "blog/index.html": "fbfa2611c97cc5426d78450078df36fb1818da1a5eeaef718e6a18fa9fd3d287","blog/japanese-speech-recognition.html": "399f5ce84e68ac00854bdf70b52b1cde6795efca7c800fd6492035a37a7c1b68","blog/lightweight-speech-recognition-cpu.html": "d6270066222ed5baef0df108e6a4155f4a87bb5482169f23f218794d371f3281","blog/punctuation-restoration-python.html": "6cd26e03bf75b4343afd7b351c2681ebd513b058fb549c5888802be9ff3a1229",@@ -54,11 +55,12 @@"en/blog/funasr-v1-3-28-realtime-websocket-subtitles.html": "233534cb4306387f1a9a695886a5f515679dff591d04b90827f41a668d7f7ada","en/blog/funasr-v1-4-0-pypi-release.html": "76fce7c27ad057d160236541369fa6c6d9f143ad03754fd3269ff2c4e32908f9","en/blog/funasr-v1-4-3-pypi-release.html": "b9007b4e74de38a4a7a38ef6f3659336c16c2159d454f290e7e702a4d59cc4d5",+ "en/blog/funasr-v1-4-5-pypi-llama-cpp-release.html": "b5944d0d3dc7f845d6e8de155958ded3f0196571a65c9e11b5b01a7f9ea7ed6b","en/blog/funasr-vs-faster-whisper-chinese.html": "abf31d3827dfba9b31ccd4e76229de9e4bfff00ab8bcee33bd718c88249630f4","en/blog/funasr-vs-whisper-benchmark.html": "367d4a8a1cc09ac932c80cdad065127925c5e5f942a806ace683c16dc1132769","en/blog/funclip-v2-1-0-video-clipping-release.html": "229baf59adf2c3290541d9b3c8a6243992406ba84b712714e9d14599205cb1d4","en/blog/generate-subtitles-srt-vtt-from-audio-video.html": "1530d4b9e94820a0b60d801e8d1c19b7aeb031b2f2b76c092657392e47706c6d",- "en/blog/index.html": "e5133a6b41024c574f4d9ae7df9d3db2094913c67de69f3ae8d5639d61acc37f",+ "en/blog/index.html": "526253689a12ead8058ef9f45b9cdc37254f6507ab6d23511a742d88fddb95c2","en/blog/japanese-speech-recognition.html": "c476adcc2be1ed7c19e2345cc91b8ee6a0e04efd794b9b6476dd5d9b3a2c8b04","en/blog/lightweight-speech-recognition-cpu.html": "0e02c0e0853b12613d32c250f593c05c2052f18231207b63ee01d6e6600a86f7","en/blog/punctuation-restoration-python.html": "d6e30049f48d9dc6e6bc22eb567830013e4161eca51215616738b675dd1ffc40",@@ -104,7 +106,7 @@"models.html": "724c88917c9934ffa66948cf3eb583fed930dd689d691b7d21fa22ee8332e606","quickstart.html": "a543bf07c2cfc9a69ef3831f15c9e9d3e571a826a73cf7e4bdc8e9eb7b910254","robots.txt": "1305b0f680b3e9f73b01301dc474423230dffab85731a379a694f30c73c07f2d",- "sitemap.xml": "967778b911c834ffeb186126353b4e8cecc009c5c73d4684b9978363e97ede34",+ "sitemap.xml": "bcc17d763cd1b0d317b3af8e52cc40fde58c4a80f3fb5664a12b5503ecf2fc19","static/liveplayer/liveplayer-component.min.js": "ea10fda35574cc3a77a3e7187095467d1a409009a2d39690051d33d0e7055d30","static/liveplayer/liveplayer-lib.min.js": "ad67b4e1188c586ec218674a6db9968daf71ed4fbc262258f061c1f931d54250","static/offline/index.html": "868781c621ddabe5d3f088d6b739e0c08e6cdc150587af6515cc323852018e62",diff --git a/web-pages/product-site/legacy/blog/funasr-v1-4-5-pypi-llama-cpp-release.html b/web-pages/product-site/legacy/blog/funasr-v1-4-5-pypi-llama-cpp-release.htmlnew file mode 100644index 000000000..099968cc4--- /dev/null+++ b/web-pages/product-site/legacy/blog/funasr-v1-4-5-pypi-llama-cpp-release.html@@ -0,0 +1,67 @@+<!DOCTYPE html>+<html lang="zh">+<head>+ <meta charset="utf-8">+ <meta name="viewport" content="width=device-width,initial-scale=1">+ <title>FunASR v1.4.5:更轻的 Python 推理与九平台 llama.cpp 运行时 | FunASR</title>+ <meta name="description" content="FunASR v1.4.5 移除 Python 推理对 torchaudio 的硬依赖,新增可选 kaldi-native-fbank,并在同一 Release 提供九个平台 llama.cpp 运行时、PyPI 包与 SHA-256 清单。">+ <meta name="keywords" content="FunASR v1.4.5,PyPI,torchaudio,kaldi-native-fbank,llama.cpp,GGUF,Ascend 910B,语音识别部署">+ <link rel="icon" href="/logo.png">+ <meta property="og:title" content="FunASR v1.4.5:更轻的 Python 推理与九平台 llama.cpp 运行时">+ <meta property="og:description" content="普通 PyPI 安装不再硬依赖 torchaudio;一个发布页覆盖 Python 包与 Linux、macOS、Windows 运行时。">+ <meta property="og:type" content="article">+ <meta property="og:url" content="https://www.funasr.com/blog/funasr-v1-4-5-pypi-llama-cpp-release.html">+ <meta property="og:image" content="https://www.funasr.com/img/banner.4f436d19.png">+ <link rel="canonical" href="https://www.funasr.com/blog/funasr-v1-4-5-pypi-llama-cpp-release.html">+ <link rel="alternate" hreflang="en" href="https://www.funasr.com/en/blog/funasr-v1-4-5-pypi-llama-cpp-release.html">+ <link rel="alternate" hreflang="zh" href="https://www.funasr.com/blog/funasr-v1-4-5-pypi-llama-cpp-release.html">+ <link rel="alternate" hreflang="x-default" href="https://www.funasr.com/en/blog/funasr-v1-4-5-pypi-llama-cpp-release.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":"BlogPosting","headline":"FunASR v1.4.5:更轻的 Python 推理与九平台 llama.cpp 运行时","description":"普通 PyPI 安装不再硬依赖 torchaudio;一个发布页覆盖 Python 包与 Linux、macOS、Windows 运行时。","datePublished":"2026-08-28","dateModified":"2026-08-28","author":{"@type":"Organization","name":"FunASR / Tongyi Lab"},"publisher":{"@type":"Organization","name":"FunASR","logo":{"@type":"ImageObject","url":"https://www.funasr.com/logo.png"}},"mainEntityOfPage":{"@type":"WebPage","@id":"https://www.funasr.com/blog/funasr-v1-4-5-pypi-llama-cpp-release.html"},"image":"https://www.funasr.com/img/banner.4f436d19.png"}+ </script>+ <style>+ :root{--primary:#2563eb;--primary-dark:#1d4ed8;--accent:#047857;--bg:#fff;--surface:#f8fafc;--border:#e2e8f0;--text:#0f172a;--text-soft:#475569;--text-muted:#94a3b8}+ *{margin:0;padding:0;box-sizing:border-box}body{font-family:'Inter',-apple-system,sans-serif;color:var(--text);line-height:1.7;background:var(--bg)}+ a{color:var(--primary);text-decoration:none}a:hover{text-decoration:underline}.container{max-width:800px;margin:0 auto;padding:0 24px}+ .nav{position:fixed;top:0;left:0;right:0;z-index:100;background:rgba(255,255,255,.94);backdrop-filter:blur(12px);border-bottom:1px solid var(--border);padding:14px 0}.nav .container{max-width:1120px;display:flex;align-items:center;gap:24px}.nav-logo{font-size:1.2rem;font-weight:800;color:var(--text)}.nav-logo span{color:var(--primary)}.nav-links{display:flex;gap:20px;margin-left:auto;flex-wrap:wrap}.nav-links a{font-size:.85rem;font-weight:500;color:var(--text-soft)}.nav-btn{padding:7px 16px;background:var(--primary);color:#fff;border-radius:7px;font-size:.83rem;font-weight:600}+ article{padding:116px 0 80px}h1{font-size:2rem;font-weight:800;line-height:1.3;margin-bottom:14px}h2{font-size:1.3rem;font-weight:700;margin:38px 0 14px;padding-top:18px;border-top:1px solid var(--border)}p{margin-bottom:14px;color:var(--text-soft)}ul{margin:0 0 16px 22px;color:var(--text-soft)}li{margin-bottom:7px}table{width:100%;border-collapse:collapse;margin:18px 0;font-size:.88rem}th,td{padding:11px 12px;text-align:left;vertical-align:top;border-bottom:1px solid var(--border)}th{font-weight:600;background:var(--surface);color:var(--text)}pre{background:#172033;color:#e2e8f0;padding:18px;border-radius:8px;overflow-x:auto;font-size:.82rem;line-height:1.7;margin:14px 0}code{font-family:ui-monospace,Menlo,monospace}p code,li code,td code{background:var(--surface);padding:2px 6px;border-radius:5px;font-size:.85em;color:var(--primary-dark)}.post-meta{color:var(--text-muted);font-size:.85rem;margin-bottom:22px}.lead{font-size:1.05rem;color:var(--text);margin:22px 0}.hero-media{display:block;width:100%;aspect-ratio:2.04/1;object-fit:cover;border-radius:8px;border:1px solid var(--border);margin:0 0 24px}.proof{border-left:4px solid var(--accent);background:#f0fdf4;padding:14px 16px;margin:18px 0;color:#14532d}.cta{background:var(--surface);border:1px solid var(--border);border-radius:8px;padding:24px;margin:30px 0;text-align:center}.cta p{margin-bottom:16px}.cta a.btn{display:inline-block;padding:10px 22px;background:var(--primary);color:#fff;border-radius:7px;font-weight:600}footer{background:#0f172a;color:#94a3b8;padding:28px 0;text-align:center;font-size:.8rem}footer a{color:#94a3b8}@media(max-width:900px){.nav-links{display:none}.nav .container{gap:16px}.nav-logo{margin-right:auto}h1{font-size:1.65rem}table{font-size:.8rem}th,td{padding:9px 8px}}+ </style>+</head>+<body>+<nav class="nav"><div class="container"><a href="/" class="nav-logo">Fun<span>ASR</span></a><div class="nav-links"><a href="/deploy/">部署</a><a href="/models.html">模型</a><a href="/benchmarks.html">性能</a><a href="/ecosystem.html">生态</a><a href="/blog/">博客</a><a href="/donors.html">功德榜</a></div><a href="/en/blog/funasr-v1-4-5-pypi-llama-cpp-release.html" style="font-size:.83rem;font-weight:500;color:var(--text-soft)">EN</a><a href="/go/github" target="_blank" rel="noopener" class="nav-btn">GitHub ★</a></div></nav>+<article><div class="container">+ <h1>FunASR v1.4.5:更轻的 Python 推理与九平台 llama.cpp 运行时</h1>+ <p class="post-meta">2026-08-28 · FunASR 正式发布</p>+ <img class="hero-media" src="/img/banner.4f436d19.png?v=1.4.5" alt="FunASR v1.4.5 发布视觉图">+ <p class="lead">FunASR v1.4.5 把 Python 推理的默认安装做得更轻:<code>torchaudio</code> 不再是硬依赖,需要原生 fbank 时可选择 <code>kaldi-native-fbank</code>。同一个 GitHub Release 还提供 wheel、sdist、九个平台的 llama.cpp / GGUF 运行时和校验清单。</p>++ <h2>安装边界</h2>+ <table><thead><tr><th>路径</th><th>命令</th><th>适用场景</th></tr></thead><tbody>+ <tr><td>标准安装</td><td><code>python -m pip install -U "funasr==1.4.5"</code></td><td>常规 Python 推理;不再自动安装 torchaudio</td></tr>+ <tr><td>KNF 后端</td><td><code>python -m pip install -U "funasr[knf]==1.4.5"</code></td><td>需要 kaldi-native-fbank 的轻量特征提取路径</td></tr>+ <tr><td>既有 torchaudio 流程</td><td><code>python -m pip install torchaudio</code></td><td>仍可显式安装并继续使用,不属于默认依赖</td></tr>+ </tbody></table>+ <p>正式包位于 <a href="https://pypi.org/project/funasr/1.4.5/" target="_blank" rel="noopener">PyPI 1.4.5</a>。wheel 与 sdist 均从精确 main 提交 <code>4df59cc1</code> 构建并回源下载校验。</p>++ <h2>部署证据与边界</h2>+ <div class="proof">Ascend 910B 的记录在既定测试契约下处理 70.47 秒音频耗时 1.15 秒,RTF 0.016。该结果是固定硬件、模型、音频和计时范围的复现实证,不代表其他设备或生产并发容量。</div>+ <p>发布契约测试 38 项全部通过;wheel 与 sdist 通过 Twine、压缩包完整性、隔离安装和 public PyPI 回源校验。签名标签精确指向 <code>4df59cc15386ff3bb10916256d807ebc5c85f81d</code>。</p>++ <h2>一个页面下载全部 12 个资产</h2>+ <p><a href="https://github.com/modelscope/FunASR/releases/tag/v1.4.5" target="_blank" rel="noopener">v1.4.5 Release</a> 固定复用 <a href="https://github.com/modelscope/FunASR/releases/tag/runtime-llamacpp-v0.2.1" target="_blank" rel="noopener">runtime-llamacpp-v0.2.1</a>,并附带:</p>+ <ul>+ <li>Python:<code>funasr-1.4.5-py3-none-any.whl</code>、<code>funasr-1.4.5.tar.gz</code></li>+ <li>Linux:<code>funasr-llamacpp-linux-arm64.tar.gz</code>、<code>funasr-llamacpp-linux-x64.tar.gz</code>、<code>funasr-llamacpp-linux-x64-avx2.tar.gz</code>、<code>funasr-llamacpp-linux-x64-vulkan.tar.gz</code></li>+ <li>macOS:<code>funasr-llamacpp-macos-arm64.tar.gz</code></li>+ <li>Windows:<code>funasr-llamacpp-windows-x64.zip</code>、<code>funasr-llamacpp-windows-x64-avx2.zip</code>、<code>funasr-llamacpp-windows-x64-cuda.zip</code>、<code>funasr-llamacpp-windows-x64-vulkan.zip</code></li>+ <li>校验:<code>SHA256SUMS-v1.4.5</code></li>+ </ul>+ <pre><code>sha256sum -c SHA256SUMS-v1.4.5</code></pre>+ <p>按硬件选择、启动命令、烟测和已知边界见<a href="/deploy/llama-cpp.html"> llama.cpp 工业部署页</a>;GPU 服务可继续参考 <a href="/deploy/vllm.html">vLLM 部署页</a>。</p>++ <div class="cta"><p>从固定版本和可校验资产开始,在目标硬件与真实业务音频上复测后再上线。</p><a class="btn" href="https://github.com/modelscope/FunASR/releases/tag/v1.4.5" target="_blank" rel="noopener">查看 v1.4.5 与全部 12 个资产</a></div>+</div></article>+<footer><div class="container">© 2024-2026 Tongyi Lab, Alibaba Group · <a href="https://github.com/modelscope/FunASR/blob/main/LICENSE">MIT License</a></div></footer>+</body>+</html>diff --git a/web-pages/product-site/legacy/blog/index.html b/web-pages/product-site/legacy/blog/index.htmlindex 549977da5..7925da337 100644--- a/web-pages/product-site/legacy/blog/index.html+++ b/web-pages/product-site/legacy/blog/index.html@@ -42,13 +42,13 @@ p code,li code{background:var(--surface);padding:2px 6px;border-radius:5px;font-.post-card p{margin:0;color:var(--text-soft);font-size:0.9rem}.post-card .date{color:var(--text-muted);font-size:0.8rem}.launch-feature{max-width:800px;margin:104px auto 0;padding:0 24px}.launch-feature a{display:grid;grid-template-columns:1fr auto;align-items:center;gap:20px;border:1px solid #bbf7d0;border-left:4px solid #047857;border-radius:8px;background:#f0fdf4;padding:20px 22px;color:var(--text)}.launch-feature a:hover{border-color:#047857;text-decoration:none}.launch-feature .date{font-size:.8rem;color:#047857;font-weight:600}.launch-feature h2{border:0;padding:0;margin:3px 0 5px;font-size:1.2rem}.launch-feature p{margin:0;font-size:.9rem}.launch-feature .action{font-weight:700;color:#1d4ed8;white-space:nowrap}-.previous-release{padding-top:16px}.previous-release .container{display:grid;grid-template-columns:1fr 1fr;gap:16px}.previous-release .post-card{margin-bottom:0}+.previous-release{padding-top:16px}.previous-release .container{display:grid;grid-template-columns:repeat(3,minmax(0,1fr));gap:16px}.previous-release .post-card{margin-bottom:0}footer{background:#0f172a;color:#94a3b8;padding:28px 0;text-align:center;font-size:0.8rem}footer a{color:#94a3b8}@media(max-width:900px){.nav-links{display:none}.nav .container{gap:16px}.nav-logo{margin-right:auto}.launch-feature a{grid-template-columns:1fr}.launch-feature .action{white-space:normal}.previous-release .container{grid-template-columns:1fr}}</style></head><body><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="https://www.funasr.com/en/blog/" style="font-size:0.83rem;font-weight:500;color:var(--text-soft)">EN</a><a href="/go/github" target="_blank" class="nav-btn">GitHub ★</a></div></nav>-<section class="launch-feature"><a href="/blog/funasr-v1-4-3-pypi-release.html"><div><span class="date">2026-08-21 · 正式 Release</span><h2>FunASR v1.4.3:可选 Silero VAD 与大规模固定 K 说话人聚类</h2><p>新增毫秒级 Silero VAD 适配器,降低已知说话人数的大规模聚类内存压力,并提供 12 个带 SHA-256 的发布资产。</p></div><span class="action">查看发布与校验指南 →</span></a></section>-<section class="previous-release"><div class="container"><a class="post-card" href="/blog/funasr-v1-4-0-pypi-release.html"><span class="date">2026-07-31 · 正式 Release</span><h2>FunASR v1.4.0:更完整的 PyPI 安装包与更安全的 AutoModel 参数</h2><p>修复 SenseVoice 与 RWKV-BAT 包数据,提前拒绝 vda_model 误拼写,并提供 12 个带 SHA-256 的发布资产。</p></a><a class="post-card" href="/blog/subtitle-edit-fun-asr-sensevoice-local-subtitles.html"><span class="date">2026-08-03 · 生态集成</span><h2>Subtitle Edit 5.2:Fun-ASR Nano / SenseVoice 本地视频字幕</h2><p>在 Windows、macOS 与 Linux 的成熟字幕工作区中直接选择本地 ASR 引擎,支持 Q4、Q8、F16。</p></a></div></section>+<section class="launch-feature"><a href="/blog/funasr-v1-4-5-pypi-llama-cpp-release.html"><div><span class="date">2026-08-28 · 正式 Release</span><h2>FunASR v1.4.5:更轻的 Python 推理与九平台 llama.cpp 运行时</h2><p>默认安装不再硬依赖 torchaudio,可选 kaldi-native-fbank,并在一个页面提供 PyPI 包、九平台运行时与 SHA-256 清单。</p></div><span class="action">查看安装与部署指南 →</span></a></section>+<section class="previous-release"><div class="container"><a class="post-card" href="/blog/funasr-v1-4-3-pypi-release.html"><span class="date">2026-08-21 · 正式 Release</span><h2>FunASR v1.4.3:可选 Silero VAD 与大规模固定 K 说话人聚类</h2><p>新增毫秒级 Silero VAD 适配器,降低已知说话人数的大规模聚类内存压力,并提供 12 个带 SHA-256 的发布资产。</p></a><a class="post-card" href="/blog/funasr-v1-4-0-pypi-release.html"><span class="date">2026-07-31 · 正式 Release</span><h2>FunASR v1.4.0:更完整的 PyPI 安装包与更安全的 AutoModel 参数</h2><p>修复 SenseVoice 与 RWKV-BAT 包数据,提前拒绝 vda_model 误拼写,并提供 12 个带 SHA-256 的发布资产。</p></a><a class="post-card" href="/blog/subtitle-edit-fun-asr-sensevoice-local-subtitles.html"><span class="date">2026-08-03 · 生态集成</span><h2>Subtitle Edit 5.2:Fun-ASR Nano / SenseVoice 本地视频字幕</h2><p>在 Windows、macOS 与 Linux 的成熟字幕工作区中直接选择本地 ASR 引擎,支持 Q4、Q8、F16。</p></a></div></section><article><div class="container"><h1>FunASR 技术博客</h1><p class="post-meta">语音识别部署、教程与最佳实践</p><a class="post-card" href="/blog/speech-to-text-timestamps-python.html"><span class="date">2026-06-23</span><h2>语音识别带时间戳(字级 timestamps)Python 实战:每个字精确到毫秒</h2><p>FunASR Paraformer 原生字级时间戳:每个字带 [起始毫秒,结束毫秒],一次调用即有。可做逐字高亮、点击跳转、字幕对齐。附真实实测+配对代码。</p></a><a class="post-card" href="/blog/punctuation-restoration-python.html"><span class="date">2026-06-23</span><h2>标点恢复 Python 实战:给无标点文本/ASR 结果自动加标点</h2><p>FunASR ct-punc 开源标点恢复,中英双语,3 行 Python 补全 ,。?(英文还做句首大写),也能一行挂到 ASR 上。附真实实测。</p></a><a class="post-card" href="/blog/chinese-speech-recognition.html"><span class="date">2026-06-22</span><h2>中文语音识别(普通话)Python 实战:用 FunASR 又快又准</h2><p>FunASR 专为中文打造:默认旗舰 Fun-ASR-Nano(CER 8.06%),CPU 用 SenseVoice(7.81%)/ Paraformer(10.18%,时间戳/热词),都远好于 Whisper(~20%)。3 行 Python。</p></a><a class="post-card" href="/blog/self-hosted-google-aws-azure-speech-to-text.html"><span class="date">2026-06-22</span><h2>自托管语音转文字:Google / AWS / Azure 云语音 API 的免费开源替代</h2><p>开源免费(MIT)、本地推理、不按分钟计费、数据不出内网、中文强,OpenAI 兼容改 base_url 即可迁移。附实测代码 + FunASR vs 云 API 对比 + 成本分析。</p></a><a class="post-card" href="/blog/voice-activity-detection-python.html"><span class="date">2026-06-22</span><h2>Python 语音活动检测(VAD):检测语音、去静音、按停顿切分音频</h2><p>FunASR fsmn-vad 3 行代码返回每段语音的起止毫秒。实测 13s 录音 0.12s 切成 2 段、去除 18% 静音;可去静音、切分长录音、给 Whisper 等做预处理防幻觉。</p></a><a class="post-card" href="/blog/japanese-speech-recognition.html"><span class="date">2026-06-22</span><h2>日语语音识别:SenseVoice 一个模型搞定日语转写+标点+情感</h2><p>同一段日语音频实测:SenseVoice 写对同音字 転売、自动加标点,Whisper-small 误为 天売、无标点。原生 ja 支持 + 自动语种识别 + 情感事件,3 行 Python。</p></a><a class="post-card" href="/blog/self-hosted-deepgram-assemblyai-alternative.html"><span class="date">2026-06-23</span><h2>自托管替代 Deepgram / AssemblyAI</h2><p>开源免费、自托管、音频不出本地、中文领先,自带 OpenAI 兼容 API——客户端只改 base_url 即可替代按分钟付费的云 STT。</p></a><a class="post-card" href="/blog/which-funasr-model.html"><span class="date">2026-06-23</span><h2>选哪个 FunASR 模型?Nano vs MLT-Nano vs SenseVoice vs Paraformer</h2><p>模型选型表 + 场景代码:中英日及中文方言/口音选旗舰 Nano,31 语种选独立 MLT-Nano,CPU 再选 SenseVoice 或 Paraformer。</p></a><a class="post-card" href="/blog/lightweight-speech-recognition-cpu.html"><span class="date">2026-06-22</span><h2>轻量语音识别:CPU 上约 250MB 跑中文 ASR</h2><p>单二进制 + 254MB q8 模型,无需 GPU/Python,CPU 0.16s,中文 CER 7.99%——比 whisper.cpp small 还小且准 3 倍。</p></a><a class="post-card" href="/blog/cantonese-speech-recognition.html"><span class="date">2026-06-21</span><h2>粤语语音识别:SenseVoice 原生支持粤语口语(Whisper 会转成普通话)</h2><p>同一段粤语音频实测:SenseVoice 保留 呢/唔/嘅,Whisper 转成普通话书面语。原生 yue 支持 + 自动语种识别,3 行 Python。</p></a><a class="post-card" href="/blog/funasr-vs-faster-whisper-chinese.html"><span class="date">2026-06-21</span><h2>FunASR vs faster-whisper:中文与粤语实测对比</h2><p>粤语被 faster-whisper 误判为普通话、日语同音字错;SenseVoice 原生支持粤语+语种识别,中文 CER 低约 2.7 倍。实测。</p></a><a class="post-card" href="/blog/funasr-llama-cpp-whisper-cpp-alternative.html"><span class="date">2026-06-20</span><h2>FunASR 跑进 llama.cpp:中文 ASR 的 whisper.cpp 替代品(CPU/零 Python)</h2><p>单个自包含二进制、内置 VAD、吃任意音频,下载即用转写中文;中文 CPU 上比 whisper.cpp 准约 2.7 倍。3 步实测。</p></a><a class="post-card" href="/blog/speech-to-text-python-transcribe-audio.html"><span class="date">2026-06-18</span><h2>Python 语音转文字:用 FunASR 本地免费转写音频</h2><p>几行 Python 把音频转成文本,带时间戳/说话人/批量;本地、免费、无 API key、中文强。</p></a><a class="post-card" href="/blog/generate-subtitles-srt-vtt-from-audio-video.html"><span class="date">2026-06-18</span><h2>用 FunASR 自动生成字幕:音频/视频一键出 SRT 和 VTT</h2><p>一行命令出 SRT,Python 同时导出 VTT,带说话人和真实时间戳;本地、免费、中文强。</p></a><a class="post-card" href="/blog/self-hosted-openai-whisper-api-alternative.html"><span class="date">2026-06-18</span><h2>自托管 OpenAI Whisper API 替代:FunASR 起兼容 /v1/audio/transcriptions 服务</h2><p>funasr-server 暴露 OpenAI 兼容接口,OpenAI SDK 只改 base_url 就能用;本地、免费、隐私、中文更准。</p></a><a class="post-card" href="/blog/funasr-cli-transcribe-command-line.html"><span class="date">2026-06-18</span><h2>用 FunASR 命令行转写音频:文本/JSON/SRT 字幕</h2><p>一行命令出文字/字幕/JSON,--spk 带说话人;还能 funasr-server 起 OpenAI 兼容 API。</p></a><a class="post-card" href="/blog/funasr-transcribe-long-audio.html"><span class="date">2026-06-17</span><h2>用 FunASR 转写超长音频:1 小时一次搞定</h2><p>Whisper 限 30 秒,FunASR 内置 VAD 一次吃下任意时长;实测 13 分钟 4.3 秒转完(186x)。</p></a><a class="post-card" href="/blog/funasr-realtime-streaming-asr.html"><span class="date">2026-06-17</span><h2>用 FunASR 实现实时流式语音识别(边说边出字)</h2><p>600ms 级低延迟流式 ASR:分块+cache 边说边出字,含 2-pass(流式+离线)最佳实践。</p></a><a class="post-card" href="/blog/sensevoice-emotion-language-detection.html"><span class="date">2026-06-17</span><h2>超越转写:用 SenseVoice 识别语言、情感与声学事件</h2><p>一次非自回归前向同时输出转写+语种+情感+音频事件,Whisper 做不到的四合一。</p></a><a class="post-card" href="/blog/funasr-speaker-diarization.html"><span class="date">2026-06-17</span><h2>用 FunASR 做说话人分离:谁在何时说了什么</h2><p>一次 generate 调用同时输出转写+说话人标签+时间戳,替代 pyannote+Whisper,无需 HF 授权。</p></a><a class="post-card" href="/blog/funasr-vs-whisper-benchmark.html"><span class="date">2026-06-16</span><h2>FunASR vs Whisper 实测对比:谁更快更准</h2><p>184 中文文件 H100 实测:SenseVoice 169.6x、CER 7.81%,完整速度+准确率数据。</p></a><a class="post-card" href="/blog/fun-asr-nano-guide.html"><span class="date">2026-06-16</span><h2>Fun-ASR-Nano 使用指南:800M 端到端语音识别大模型</h2><p>主力旗舰,中英日 + 7 大中文方言/26 种口音,热词/流式/说话人分离;31 语种请用 MLT-Nano。</p></a><a class="post-card" href="/blog/sensevoice-deployment-guide.html"><span class="date">2026-06-16</span><h2>SenseVoice 部署指南:五语种识别、情感与音频事件</h2><p>3 行代码跑通多语言识别,含语种/情感/事件检测、VAD、GPU/CPU。</p></a><p>更多:<a href="/quickstart.html">快速上手</a> · <a href="/models.html">模型</a></p></div></article><footer><div class="container">© 2024-2026 Tongyi Lab, Alibaba Group · <a href="https://github.com/modelscope/FunASR/blob/main/LICENSE">MIT License</a></div></footer></body></html>diff --git a/web-pages/product-site/legacy/en/blog/funasr-v1-4-5-pypi-llama-cpp-release.html b/web-pages/product-site/legacy/en/blog/funasr-v1-4-5-pypi-llama-cpp-release.htmlnew file mode 100644index 000000000..a189336ab--- /dev/null+++ b/web-pages/product-site/legacy/en/blog/funasr-v1-4-5-pypi-llama-cpp-release.html@@ -0,0 +1,67 @@+<!DOCTYPE html>+<html lang="en">+<head>+ <meta charset="utf-8">+ <meta name="viewport" content="width=device-width,initial-scale=1">+ <title>FunASR v1.4.5: Lighter Python Inference and Nine llama.cpp Runtimes | FunASR</title>+ <meta name="description" content="FunASR v1.4.5 removes the hard torchaudio inference dependency, adds optional kaldi-native-fbank, and ships PyPI packages with nine verified llama.cpp runtimes and SHA-256 checksums.">+ <meta name="keywords" content="FunASR v1.4.5,PyPI,torchaudio,kaldi-native-fbank,llama.cpp,GGUF,Ascend 910B,speech recognition deployment">+ <link rel="icon" href="/logo.png">+ <meta property="og:title" content="FunASR v1.4.5: Lighter Python Inference and Nine llama.cpp Runtimes">+ <meta property="og:description" content="The default PyPI install no longer requires torchaudio; one release page covers Python plus Linux, macOS, and Windows runtimes.">+ <meta property="og:type" content="article">+ <meta property="og:url" content="https://www.funasr.com/en/blog/funasr-v1-4-5-pypi-llama-cpp-release.html">+ <meta property="og:image" content="https://www.funasr.com/img/banner.4f436d19.png">+ <link rel="canonical" href="https://www.funasr.com/en/blog/funasr-v1-4-5-pypi-llama-cpp-release.html">+ <link rel="alternate" hreflang="en" href="https://www.funasr.com/en/blog/funasr-v1-4-5-pypi-llama-cpp-release.html">+ <link rel="alternate" hreflang="zh" href="https://www.funasr.com/blog/funasr-v1-4-5-pypi-llama-cpp-release.html">+ <link rel="alternate" hreflang="x-default" href="https://www.funasr.com/en/blog/funasr-v1-4-5-pypi-llama-cpp-release.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":"BlogPosting","headline":"FunASR v1.4.5: Lighter Python Inference and Nine llama.cpp Runtimes","description":"The default PyPI install no longer requires torchaudio; one release page covers Python plus Linux, macOS, and Windows runtimes.","datePublished":"2026-08-28","dateModified":"2026-08-28","author":{"@type":"Organization","name":"FunASR / Tongyi Lab"},"publisher":{"@type":"Organization","name":"FunASR","logo":{"@type":"ImageObject","url":"https://www.funasr.com/logo.png"}},"mainEntityOfPage":{"@type":"WebPage","@id":"https://www.funasr.com/en/blog/funasr-v1-4-5-pypi-llama-cpp-release.html"},"image":"https://www.funasr.com/img/banner.4f436d19.png"}+ </script>+ <style>+ :root{--primary:#2563eb;--primary-dark:#1d4ed8;--accent:#047857;--bg:#fff;--surface:#f8fafc;--border:#e2e8f0;--text:#0f172a;--text-soft:#475569;--text-muted:#94a3b8}+ *{margin:0;padding:0;box-sizing:border-box}body{font-family:'Inter',-apple-system,sans-serif;color:var(--text);line-height:1.7;background:var(--bg)}+ a{color:var(--primary);text-decoration:none}a:hover{text-decoration:underline}.container{max-width:800px;margin:0 auto;padding:0 24px}+ .nav{position:fixed;top:0;left:0;right:0;z-index:100;background:rgba(255,255,255,.94);backdrop-filter:blur(12px);border-bottom:1px solid var(--border);padding:14px 0}.nav .container{max-width:1120px;display:flex;align-items:center;gap:24px}.nav-logo{font-size:1.2rem;font-weight:800;color:var(--text)}.nav-logo span{color:var(--primary)}.nav-links{display:flex;gap:20px;margin-left:auto;flex-wrap:wrap}.nav-links a{font-size:.85rem;font-weight:500;color:var(--text-soft)}.nav-btn{padding:7px 16px;background:var(--primary);color:#fff;border-radius:7px;font-size:.83rem;font-weight:600}+ article{padding:116px 0 80px}h1{font-size:2rem;font-weight:800;line-height:1.3;margin-bottom:14px}h2{font-size:1.3rem;font-weight:700;margin:38px 0 14px;padding-top:18px;border-top:1px solid var(--border)}p{margin-bottom:14px;color:var(--text-soft)}ul{margin:0 0 16px 22px;color:var(--text-soft)}li{margin-bottom:7px}table{width:100%;border-collapse:collapse;margin:18px 0;font-size:.88rem}th,td{padding:11px 12px;text-align:left;vertical-align:top;border-bottom:1px solid var(--border)}th{font-weight:600;background:var(--surface);color:var(--text)}pre{background:#172033;color:#e2e8f0;padding:18px;border-radius:8px;overflow-x:auto;font-size:.82rem;line-height:1.7;margin:14px 0}code{font-family:ui-monospace,Menlo,monospace}p code,li code,td code{background:var(--surface);padding:2px 6px;border-radius:5px;font-size:.85em;color:var(--primary-dark)}.post-meta{color:var(--text-muted);font-size:.85rem;margin-bottom:22px}.lead{font-size:1.05rem;color:var(--text);margin:22px 0}.hero-media{display:block;width:100%;aspect-ratio:2.04/1;object-fit:cover;border-radius:8px;border:1px solid var(--border);margin:0 0 24px}.proof{border-left:4px solid var(--accent);background:#f0fdf4;padding:14px 16px;margin:18px 0;color:#14532d}.cta{background:var(--surface);border:1px solid var(--border);border-radius:8px;padding:24px;margin:30px 0;text-align:center}.cta p{margin-bottom:16px}.cta a.btn{display:inline-block;padding:10px 22px;background:var(--primary);color:#fff;border-radius:7px;font-weight:600}footer{background:#0f172a;color:#94a3b8;padding:28px 0;text-align:center;font-size:.8rem}footer a{color:#94a3b8}@media(max-width:900px){.nav-links{display:none}.nav .container{gap:16px}.nav-logo{margin-right:auto}h1{font-size:1.65rem}table{font-size:.8rem}th,td{padding:9px 8px}}+ </style>+</head>+<body>+<nav class="nav"><div class="container"><a href="/en/" class="nav-logo">Fun<span>ASR</span></a><div class="nav-links"><a href="/en/deploy/">Deploy</a><a href="/en/models.html">Models</a><a href="/en/benchmarks.html">Benchmarks</a><a href="/en/ecosystem.html">Ecosystem</a><a href="/en/blog/">Blog</a><a href="/en/donors.html">Donors</a></div><a href="/blog/funasr-v1-4-5-pypi-llama-cpp-release.html" style="font-size:.83rem;font-weight:500;color:var(--text-soft)">中文</a><a href="/go/github" target="_blank" rel="noopener" class="nav-btn">GitHub ★</a></div></nav>+<article><div class="container">+ <h1>FunASR v1.4.5: Lighter Python Inference and Nine llama.cpp Runtimes</h1>+ <p class="post-meta">August 28, 2026 · Stable FunASR release</p>+ <img class="hero-media" src="/img/banner.4f436d19.png?v=1.4.5" alt="FunASR v1.4.5 release visual">+ <p class="lead">FunASR v1.4.5 makes the default Python inference install lighter: <code>torchaudio</code> is no longer a hard dependency, while <code>kaldi-native-fbank</code> is available when a native fbank path is needed. The same GitHub Release carries the wheel, sdist, nine llama.cpp / GGUF runtimes, and one checksum manifest.</p>++ <h2>Installation boundaries</h2>+ <table><thead><tr><th>Path</th><th>Command</th><th>Use it for</th></tr></thead><tbody>+ <tr><td>Standard</td><td><code>python -m pip install -U "funasr==1.4.5"</code></td><td>Normal Python inference without an automatic torchaudio install</td></tr>+ <tr><td>KNF backend</td><td><code>python -m pip install -U "funasr[knf]==1.4.5"</code></td><td>A lighter feature path backed by kaldi-native-fbank</td></tr>+ <tr><td>Existing torchaudio flow</td><td><code>python -m pip install torchaudio</code></td><td>Still supported when installed explicitly; it is outside the default dependency set</td></tr>+ </tbody></table>+ <p>The stable distributions are on <a href="https://pypi.org/project/funasr/1.4.5/" target="_blank" rel="noopener">PyPI 1.4.5</a>. Both were built from exact main commit <code>4df59cc1</code>, downloaded again, and matched to their published digests.</p>++ <h2>Deployment evidence and limits</h2>+ <div class="proof">The recorded Ascend 910B run processed 70.47 seconds of audio in 1.15 seconds under its fixed test contract, for RTF 0.016. This is reproducible evidence for that hardware, model, audio, and timing scope, not a production capacity promise for other systems.</div>+ <p>All 38 release-contract tests passed. The wheel and sdist passed Twine, archive integrity, isolated-install, and public-PyPI round-trip checks. The signed tag resolves exactly to <code>4df59cc15386ff3bb10916256d807ebc5c85f81d</code>.</p>++ <h2>All 12 assets on one page</h2>+ <p>The <a href="https://github.com/modelscope/FunASR/releases/tag/v1.4.5" target="_blank" rel="noopener">v1.4.5 Release</a> pins and mirrors <a href="https://github.com/modelscope/FunASR/releases/tag/runtime-llamacpp-v0.2.1" target="_blank" rel="noopener">runtime-llamacpp-v0.2.1</a> with:</p>+ <ul>+ <li>Python: <code>funasr-1.4.5-py3-none-any.whl</code> and <code>funasr-1.4.5.tar.gz</code></li>+ <li>Linux: <code>funasr-llamacpp-linux-arm64.tar.gz</code>, <code>funasr-llamacpp-linux-x64.tar.gz</code>, <code>funasr-llamacpp-linux-x64-avx2.tar.gz</code>, <code>funasr-llamacpp-linux-x64-vulkan.tar.gz</code></li>+ <li>macOS: <code>funasr-llamacpp-macos-arm64.tar.gz</code></li>+ <li>Windows: <code>funasr-llamacpp-windows-x64.zip</code>, <code>funasr-llamacpp-windows-x64-avx2.zip</code>, <code>funasr-llamacpp-windows-x64-cuda.zip</code>, <code>funasr-llamacpp-windows-x64-vulkan.zip</code></li>+ <li>Verification: <code>SHA256SUMS-v1.4.5</code></li>+ </ul>+ <pre><code>sha256sum -c SHA256SUMS-v1.4.5</code></pre>+ <p>Use the <a href="/en/deploy/llama-cpp.html">llama.cpp production deployment page</a> for hardware selection, launch commands, smoke tests, and known boundaries. GPU API deployments can continue with the <a href="/en/deploy/vllm.html">vLLM guide</a>.</p>++ <div class="cta"><p>Start from a pinned version and verified assets, then revalidate on the target hardware with representative audio before rollout.</p><a class="btn" href="https://github.com/modelscope/FunASR/releases/tag/v1.4.5" target="_blank" rel="noopener">Open v1.4.5 and all 12 assets</a></div>+</div></article>+<footer><div class="container">© 2024-2026 Tongyi Lab, Alibaba Group · <a href="https://github.com/modelscope/FunASR/blob/main/LICENSE">MIT License</a></div></footer>+</body>+</html>diff --git a/web-pages/product-site/legacy/en/blog/index.html b/web-pages/product-site/legacy/en/blog/index.htmlindex 5f4a8668a..fe315c379 100644--- a/web-pages/product-site/legacy/en/blog/index.html+++ b/web-pages/product-site/legacy/en/blog/index.html@@ -42,13 +42,13 @@ p code,li code{background:var(--surface);padding:2px 6px;border-radius:5px;font-.post-card p{margin:0;color:var(--text-soft);font-size:0.9rem}.post-card .date{color:var(--text-muted);font-size:0.8rem}.launch-feature{max-width:800px;margin:104px auto 0;padding:0 24px}.launch-feature a{display:grid;grid-template-columns:1fr auto;align-items:center;gap:20px;border:1px solid #bbf7d0;border-left:4px solid #047857;border-radius:8px;background:#f0fdf4;padding:20px 22px;color:var(--text)}.launch-feature a:hover{border-color:#047857;text-decoration:none}.launch-feature .date{font-size:.8rem;color:#047857;font-weight:600}.launch-feature h2{border:0;padding:0;margin:3px 0 5px;font-size:1.2rem}.launch-feature p{margin:0;font-size:.9rem}.launch-feature .action{font-weight:700;color:#1d4ed8;white-space:nowrap}-.previous-release{padding-top:16px}.previous-release .container{display:grid;grid-template-columns:1fr 1fr;gap:16px}.previous-release .post-card{margin-bottom:0}+.previous-release{padding-top:16px}.previous-release .container{display:grid;grid-template-columns:repeat(3,minmax(0,1fr));gap:16px}.previous-release .post-card{margin-bottom:0}footer{background:#0f172a;color:#94a3b8;padding:28px 0;text-align:center;font-size:0.8rem}footer a{color:#94a3b8}@media(max-width:900px){.nav-links{display:none}.nav .container{gap:16px}.nav-logo{margin-right:auto}.launch-feature a{grid-template-columns:1fr}.launch-feature .action{white-space:normal}.previous-release .container{grid-template-columns:1fr}}</style></head><body><nav class="nav"><div class="container"><a href="/en/" class="nav-logo">Fun<span>ASR</span></a><div class="nav-links"><a href="/en/#capabilities">Capabilities</a><a href="/en/#usage">How to Use</a><a href="/en/#benchmark">Benchmark</a><a href="/en/#demo">Demo</a><a href="https://modelscope.github.io/FunASR/" target="_blank">Docs</a><a href="/en/quickstart.html">Quickstart</a><a href="/en/models.html">Models</a><a href="/en/ecosystem.html">Ecosystem</a><a href="/en/blog/">Blog</a><a href="/en/vs-whisper.html">vs Whisper</a><a href="/en/donors.html">Thanks</a></div><a href="https://www.funasr.com/blog/" style="font-size:0.83rem;font-weight:500;color:var(--text-soft)">中文</a><a href="/go/github" target="_blank" class="nav-btn">GitHub ★</a></div></nav>-<section class="launch-feature"><a href="/en/blog/funasr-v1-4-3-pypi-release.html"><div><span class="date">August 21, 2026 · Stable release</span><h2>FunASR v1.4.3: Optional Silero VAD and Fixed-K Speaker Clustering</h2><p>Millisecond Silero VAD segments, lower-memory clustering for large known-speaker workloads, and 12 SHA-256-addressed release assets.</p></div><span class="action">Read the release and verification guide →</span></a></section>-<section class="previous-release"><div class="container"><a class="post-card" href="/en/blog/funasr-v1-4-0-pypi-release.html"><span class="date">July 31, 2026 · Stable release</span><h2>FunASR v1.4.0: Complete PyPI Packages and Safer AutoModel Arguments</h2><p>SenseVoice and RWKV-BAT package-data fixes, an early vda_model typo guard, and 12 SHA-256-addressed release assets.</p></a><a class="post-card" href="/en/blog/subtitle-edit-fun-asr-sensevoice-local-subtitles.html"><span class="date">August 3, 2026 · Ecosystem</span><h2>Subtitle Edit 5.2: Local Video Subtitles with Fun-ASR Nano or SenseVoice</h2><p>Select a local ASR engine inside a mature subtitle workspace on Windows, macOS, or Linux, with Q4, Q8, and F16 models.</p></a></div></section>+<section class="launch-feature"><a href="/en/blog/funasr-v1-4-5-pypi-llama-cpp-release.html"><div><span class="date">August 28, 2026 · Stable release</span><h2>FunASR v1.4.5: Lighter Python Inference and Nine llama.cpp Runtimes</h2><p>No hard torchaudio dependency in the default install, optional kaldi-native-fbank, and one page for PyPI packages, nine runtimes, and SHA-256 checksums.</p></div><span class="action">Read the install and deployment guide →</span></a></section>+<section class="previous-release"><div class="container"><a class="post-card" href="/en/blog/funasr-v1-4-3-pypi-release.html"><span class="date">August 21, 2026 · Stable release</span><h2>FunASR v1.4.3: Optional Silero VAD and Fixed-K Speaker Clustering</h2><p>Millisecond Silero VAD segments, lower-memory clustering for large known-speaker workloads, and 12 SHA-256-addressed release assets.</p></a><a class="post-card" href="/en/blog/funasr-v1-4-0-pypi-release.html"><span class="date">July 31, 2026 · Stable release</span><h2>FunASR v1.4.0: Complete PyPI Packages and Safer AutoModel Arguments</h2><p>SenseVoice and RWKV-BAT package-data fixes, an early vda_model typo guard, and 12 SHA-256-addressed release assets.</p></a><a class="post-card" href="/en/blog/subtitle-edit-fun-asr-sensevoice-local-subtitles.html"><span class="date">August 3, 2026 · Ecosystem</span><h2>Subtitle Edit 5.2: Local Video Subtitles with Fun-ASR Nano or SenseVoice</h2><p>Select a local ASR engine inside a mature subtitle workspace on Windows, macOS, or Linux, with Q4, Q8, and F16 models.</p></a></div></section><article><div class="container"><h1>FunASR Blog</h1><p class="post-meta">Speech recognition deployment, tutorials and best practices</p><a class="post-card" href="/en/blog/speech-to-text-timestamps-python.html"><span class="date">2026-06-23</span><h2>Speech-to-Text with Word/Character-Level Timestamps in Python</h2><p>FunASR Paraformer gives native character timestamps: every char has [start_ms, end_ms] in one call. Build word highlighting, click-to-seek, subtitle alignment. Real output + pairing code.</p></a><a class="post-card" href="/en/blog/punctuation-restoration-python.html"><span class="date">2026-06-23</span><h2>Punctuation Restoration in Python — Add Punctuation to Text / ASR Output</h2><p>FunASR ct-punc: open-source bilingual (zh+en) punctuation restoration. 3 lines of Python to add ,。? (and capitalize English), or attach to ASR in one line. Real output.</p></a><a class="post-card" href="/en/blog/chinese-speech-recognition.html"><span class="date">2026-06-22</span><h2>Chinese (Mandarin) Speech Recognition in Python — Fast & Accurate with FunASR</h2><p>Purpose-built for Chinese: default flagship Fun-ASR-Nano (CER 8.06%), with SenseVoice (7.81%) / Paraformer (10.18%, timestamps/hotwords) for CPU — all far better than Whisper (~20%).</p></a><a class="post-card" href="/en/blog/self-hosted-google-aws-azure-speech-to-text.html"><span class="date">2026-06-22</span><h2>Self-Hosted Speech-to-Text — Free Alternative to Google / AWS / Azure Cloud Speech APIs</h2><p>Open-source (MIT), local inference, no per-minute billing, audio stays on your network, strong Chinese, OpenAI-compatible (migrate via base_url). Runnable code + FunASR vs cloud comparison + cost analysis.</p></a><a class="post-card" href="/en/blog/voice-activity-detection-python.html"><span class="date">2026-06-22</span><h2>Voice Activity Detection in Python — Detect Speech, Remove Silence, Split Audio</h2><p>FunASR fsmn-vad returns millisecond speech spans in 3 lines. Measured: a 13s clip split into 2 segments in 0.12s, 18% silence removed. Trim silence, split long audio, preprocess Whisper to cut hallucinations.</p></a><a class="post-card" href="/en/blog/japanese-speech-recognition.html"><span class="date">2026-06-22</span><h2>Japanese Speech Recognition in Python — SenseVoice: Transcription + Punctuation + Emotion in One Pass</h2><p>Real side-by-side on one Japanese clip: SenseVoice writes 転売 correctly and adds punctuation; Whisper-small gives 天売 with none. Native ja support + auto language ID + emotion/events, 3 lines of Python.</p></a><a class="post-card" href="/en/blog/self-hosted-deepgram-assemblyai-alternative.html"><span class="date">2026-06-23</span><h2>Self-Hosted Deepgram / AssemblyAI Alternative</h2><p>Open-source, free, self-hosted, audio stays local, Chinese-leading, with an OpenAI-compatible API — replace per-minute cloud STT by changing base_url.</p></a><a class="post-card" href="/en/blog/which-funasr-model.html"><span class="date">2026-06-23</span><h2>Which FunASR Model? Nano vs MLT-Nano vs SenseVoice vs Paraformer</h2><p>A decision table with code: flagship Nano for zh/en/ja plus Chinese dialects/accents, separate MLT-Nano for 31 languages, and CPU choices SenseVoice or Paraformer.</p></a><a class="post-card" href="/en/blog/lightweight-speech-recognition-cpu.html"><span class="date">2026-06-22</span><h2>Lightweight Speech Recognition: Chinese ASR in ~250MB on CPU</h2><p>One binary + a 254MB q8 model, no GPU/Python, 0.16s on CPU, 7.99% CER — smaller than whisper.cpp small and ~3x more accurate.</p></a><a class="post-card" href="/en/blog/cantonese-speech-recognition.html"><span class="date">2026-06-21</span><h2>Cantonese Speech Recognition in Python — SenseVoice Keeps Real Cantonese (Whisper Doesn't)</h2><p>Real side-by-side on one Cantonese clip: SenseVoice keeps 呢/唔/嘅, Whisper rewrites to Mandarin. Native yue support + auto language ID, 3 lines of Python.</p></a><a class="post-card" href="/en/blog/funasr-vs-faster-whisper-chinese.html"><span class="date">2026-06-21</span><h2>FunASR vs faster-whisper: Chinese & Cantonese Compared</h2><p>faster-whisper mislabels Cantonese as Mandarin + Japanese homophone errors; SenseVoice handles Cantonese natively, ~2.7x lower CER on Chinese.</p></a><a class="post-card" href="/en/blog/funasr-llama-cpp-whisper-cpp-alternative.html"><span class="date">2026-06-20</span><h2>FunASR on llama.cpp — a whisper.cpp Alternative for Chinese ASR (CPU, no Python)</h2><p>One self-contained binary, built-in VAD, any audio — download and transcribe Chinese; ~2.7x more accurate than whisper.cpp on CPU.</p></a><a class="post-card" href="/en/blog/speech-to-text-python-transcribe-audio.html"><span class="date">2026-06-18</span><h2>Speech to Text in Python with FunASR</h2><p>Transcribe audio in a few lines of Python — timestamps, speakers, batching. Local, free, no API key.</p></a><a class="post-card" href="/en/blog/generate-subtitles-srt-vtt-from-audio-video.html"><span class="date">2026-06-18</span><h2>Auto-Generate Subtitles (SRT & VTT) from Audio or Video with FunASR</h2><p>One command for SRT, Python for VTT too, with speaker labels and real timestamps. Local, free, strong on Chinese.</p></a><a class="post-card" href="/en/blog/self-hosted-openai-whisper-api-alternative.html"><span class="date">2026-06-18</span><h2>Self-Hosted OpenAI Whisper API Alternative with FunASR</h2><p>funasr-server exposes an OpenAI-compatible /v1/audio/transcriptions; the OpenAI SDK works by changing only base_url. Local, free, private.</p></a><a class="post-card" href="/en/blog/funasr-cli-transcribe-command-line.html"><span class="date">2026-06-18</span><h2>Transcribe Audio from the Command Line with FunASR</h2><p>One command -> text/SRT/JSON, --spk for speakers; plus funasr-server for an OpenAI-compatible API.</p></a><a class="post-card" href="/en/blog/funasr-transcribe-long-audio.html"><span class="date">2026-06-17</span><h2>Transcribe Long Audio with FunASR: Hours in One Call</h2><p>Whisper caps at 30s; FunASR ingests any length via built-in VAD - 13 min in 4.3s (186x).</p></a><a class="post-card" href="/en/blog/funasr-realtime-streaming-asr.html"><span class="date">2026-06-17</span><h2>Real-Time Streaming Speech-to-Text with FunASR</h2><p>~600ms low-latency streaming ASR with chunks + cache, plus the 2-pass (streaming + offline) best practice.</p></a><a class="post-card" href="/en/blog/sensevoice-emotion-language-detection.html"><span class="date">2026-06-17</span><h2>Beyond Transcription: Language, Emotion & Audio Events with SenseVoice</h2><p>Transcription + language + emotion + audio events in one pass — the 4-in-1 Whisper cannot do.</p></a><a class="post-card" href="/en/blog/funasr-speaker-diarization.html"><span class="date">2026-06-17</span><h2>Speaker Diarization with FunASR: Who Spoke When</h2><p>Transcription + speaker labels + timestamps in one generate() call. A pyannote+Whisper alternative, no HF gated access.</p></a><a class="post-card" href="/en/blog/funasr-vs-whisper-benchmark.html"><span class="date">2026-06-16</span><h2>FunASR vs Whisper: Real Chinese ASR Benchmark</h2><p>Measured on 184 Chinese files (H100): SenseVoice 169.6x, 7.81% CER — full speed & accuracy data.</p></a><a class="post-card" href="/en/blog/fun-asr-nano-guide.html"><span class="date">2026-06-16</span><h2>Fun-ASR-Nano Guide: 800M End-to-End ASR LLM</h2><p>Flagship for zh/en/ja plus 7 Chinese dialect groups and 26 accents; choose MLT-Nano for 31 languages.</p></a><a class="post-card" href="/en/blog/sensevoice-deployment-guide.html"><span class="date">2026-06-16</span><h2>SenseVoice Deployment Guide: Five-Language ASR, Emotion and Audio Events</h2><p>Multilingual ASR in 3 lines, with language/emotion/event tags, VAD, GPU/CPU.</p></a><p>More: <a href="/en/quickstart.html">Quickstart</a> · <a href="/en/models.html">Models</a></p></div></article><footer><div class="container">© 2024-2026 Tongyi Lab, Alibaba Group · <a href="https://github.com/modelscope/FunASR/blob/main/LICENSE">MIT License</a></div></footer></body></html>diff --git a/web-pages/product-site/legacy/sitemap.xml b/web-pages/product-site/legacy/sitemap.xmlindex 1592e4f2d..ae14af506 100644--- a/web-pages/product-site/legacy/sitemap.xml+++ b/web-pages/product-site/legacy/sitemap.xml@@ -12,6 +12,8 @@<url><loc>https://www.funasr.com/en/ecosystem.html</loc><priority>0.8</priority><changefreq>weekly</changefreq></url><url><loc>https://www.funasr.com/blog/</loc><priority>0.9</priority><changefreq>weekly</changefreq></url><url><loc>https://www.funasr.com/en/blog/</loc><priority>0.9</priority><changefreq>weekly</changefreq></url>+ <url><loc>https://www.funasr.com/blog/funasr-v1-4-5-pypi-llama-cpp-release.html</loc><lastmod>2026-08-28</lastmod><priority>0.9</priority><changefreq>monthly</changefreq></url>+ <url><loc>https://www.funasr.com/en/blog/funasr-v1-4-5-pypi-llama-cpp-release.html</loc><lastmod>2026-08-28</lastmod><priority>0.9</priority><changefreq>monthly</changefreq></url><url><loc>https://www.funasr.com/blog/funasr-v1-4-0-pypi-release.html</loc><lastmod>2026-07-31</lastmod><priority>0.9</priority><changefreq>monthly</changefreq></url><url><loc>https://www.funasr.com/en/blog/funasr-v1-4-0-pypi-release.html</loc><lastmod>2026-07-31</lastmod><priority>0.9</priority><changefreq>monthly</changefreq></url><url><loc>https://www.funasr.com/blog/funclip-v2-1-0-video-clipping-release.html</loc><lastmod>2026-07-24</lastmod><priority>0.9</priority><changefreq>monthly</changefreq></url>diff --git a/web-pages/product-site/tests/browser/product-site.spec.ts b/web-pages/product-site/tests/browser/product-site.spec.tsindex c13a8882c..b93d80915 100644--- a/web-pages/product-site/tests/browser/product-site.spec.ts+++ b/web-pages/product-site/tests/browser/product-site.spec.ts@@ -300,6 +300,64 @@ test('reduced motion disables smooth scrolling', async ({ page }) => {expect(behavior).toBe('auto');});+for (const viewport of [+ { name: 'mobile', width: 390, height: 844 },+ { name: 'desktop', width: 1440, height: 900 },+]) {+ test(`v1.4.5 release discovery is stable at ${viewport.name}`, async ({ page }, testInfo) => {+ await page.setViewportSize(viewport);+ for (const release of [+ {+ language: 'zh',+ index: '/blog/',+ article: '/blog/funasr-v1-4-5-pypi-llama-cpp-release.html',+ },+ {+ language: 'en',+ index: '/en/blog/',+ article: '/en/blog/funasr-v1-4-5-pypi-llama-cpp-release.html',+ },+ ]) {+ await page.goto(release.index);+ await expect(+ page.locator(`.launch-feature a[href="${release.article}"]`),+ ).toBeVisible();+ const history = page.locator('.previous-release .post-card');+ await expect(history).toHaveCount(3);+ const indexLayout = await history.evaluateAll((cards) => ({+ overflow: document.documentElement.scrollWidth - document.documentElement.clientWidth,+ rows: new Set(cards.map((card) => Math.round(card.getBoundingClientRect().top))).size,+ }));+ expect(indexLayout.overflow).toBeLessThanOrEqual(1);+ expect(indexLayout.rows).toBe(viewport.name === 'mobile' ? 3 : 1);++ await page.goto(release.article);+ await expect(page.locator('h1')).toContainText('FunASR v1.4.5');+ await expect(page.getByText('funasr[knf]==1.4.5', { exact: false })).toBeVisible();+ await expect(page.getByText('runtime-llamacpp-v0.2.1', { exact: false })).toBeVisible();++ const articleLayout = await page.evaluate(() => {+ const navigation = document.querySelector<HTMLElement>('nav.nav');+ const heading = document.querySelector<HTMLElement>('h1');+ return {+ overflow: document.documentElement.scrollWidth - document.documentElement.clientWidth,+ navigationBottom: navigation?.getBoundingClientRect().bottom ?? 0,+ headingTop: heading?.getBoundingClientRect().top ?? 0,+ };+ });+ expect(articleLayout.overflow).toBeLessThanOrEqual(1);+ expect(articleLayout.headingTop).toBeGreaterThanOrEqual(articleLayout.navigationBottom + 16);++ await page.screenshot({+ path: testInfo.outputPath(+ `v1.4.5-release-${release.language}-${viewport.name}.png`,+ ),+ fullPage: true,+ });+ }+ });+}+test('llama.cpp blog heading clears fixed navigation on mobile', async ({ page }, testInfo) => {await page.setViewportSize({ width: 390, height: 844 });await page.goto('/blog/funasr-llama-cpp-whisper-cpp-alternative.html');diff --git a/web-pages/product-site/tests/test_legacy.py b/web-pages/product-site/tests/test_legacy.pyindex c9363f7ef..327c4c195 100644--- a/web-pages/product-site/tests/test_legacy.py+++ b/web-pages/product-site/tests/test_legacy.py@@ -286,6 +286,80 @@ def test_v140_release_pages_are_bilingual_indexed_and_precise():assert f'https://www.funasr.com/en/blog/{slug}' in sitemap+def test_v145_release_pages_are_bilingual_indexed_and_operational():+ slug = 'funasr-v1-4-5-pypi-llama-cpp-release.html'+ pages = {+ 'zh': LEGACY / 'blog' / slug,+ 'en': LEGACY / 'en' / 'blog' / slug,+ }+ release_assets = (+ 'funasr-1.4.5-py3-none-any.whl',+ 'funasr-1.4.5.tar.gz',+ 'funasr-llamacpp-linux-arm64.tar.gz',+ 'funasr-llamacpp-linux-x64-avx2.tar.gz',+ 'funasr-llamacpp-linux-x64-vulkan.tar.gz',+ 'funasr-llamacpp-linux-x64.tar.gz',+ 'funasr-llamacpp-macos-arm64.tar.gz',+ 'funasr-llamacpp-windows-x64-avx2.zip',+ 'funasr-llamacpp-windows-x64-cuda.zip',+ 'funasr-llamacpp-windows-x64-vulkan.zip',+ 'funasr-llamacpp-windows-x64.zip',+ 'SHA256SUMS-v1.4.5',+ )+ language_contracts = {+ 'zh': ('不再是硬依赖', '签名标签', '不代表其他设备或生产并发容量'),+ 'en': ('no longer a hard dependency', 'signed tag', 'not a production capacity promise'),+ }++ for language, path in pages.items():+ text = path.read_text(encoding='utf-8')+ soup = BeautifulSoup(text, 'html.parser')++ assert 'funasr==1.4.5' in text+ assert 'funasr[knf]==1.4.5' in text+ assert 'https://pypi.org/project/funasr/1.4.5/' in text+ assert 'https://github.com/modelscope/FunASR/releases/tag/v1.4.5' in text+ assert 'runtime-llamacpp-v0.2.1' in text+ assert 'SHA256SUMS-v1.4.5' in text+ assert 'torchaudio' in text+ assert 'Ascend 910B' in text+ assert all(value in text for value in ('70.47', '1.15', '0.016'))+ assert '4df59cc15386ff3bb10916256d807ebc5c85f81d' in text+ assert all(marker in text for marker in language_contracts[language])++ asset_codes = {+ code.get_text(strip=True)+ for code in soup.select('code')+ if code.get_text(strip=True).startswith(('funasr-', 'SHA256SUMS-'))+ }+ assert asset_codes == set(release_assets)++ expected_route = f'/{"" if language == "zh" else "en/"}blog/{slug}'+ assert soup.select_one('link[rel="canonical"]')['href'].endswith(expected_route)+ peer_route = f'/{"en/" if language == "zh" else ""}blog/{slug}'+ peer_language = 'en' if language == 'zh' else 'zh'+ assert soup.select_one(+ f'link[rel="alternate"][hreflang="{peer_language}"]'+ f'[href="https://www.funasr.com{peer_route}"]'+ )+ assert soup.select_one(+ 'link[rel="alternate"][hreflang="x-default"]'+ f'[href="https://www.funasr.com/en/blog/{slug}"]'+ )+ metadata = json.loads(soup.select_one('script[type="application/ld+json"]').string)+ assert metadata['datePublished'] == '2026-08-28'+ assert metadata['dateModified'] == '2026-08-28'++ zh_index = (LEGACY / 'blog' / 'index.html').read_text(encoding='utf-8')+ en_index = (LEGACY / 'en' / 'blog' / 'index.html').read_text(encoding='utf-8')+ sitemap = (LEGACY / 'sitemap.xml').read_text(encoding='utf-8')++ assert f'/blog/{slug}' in zh_index+ assert f'/en/blog/{slug}' in en_index+ assert f'https://www.funasr.com/blog/{slug}' in sitemap+ assert f'https://www.funasr.com/en/blog/{slug}' in sitemap++def test_llama_cpp_blog_points_directly_to_current_runtime_release():pages = {'zh': LEGACY / 'blog' / 'funasr-llama-cpp-whisper-cpp-alternative.html',diff --git a/web-pages/product-site/tests/test_output.py b/web-pages/product-site/tests/test_output.pyindex 2b0fca34a..8cc682bab 100644--- a/web-pages/product-site/tests/test_output.py+++ b/web-pages/product-site/tests/test_output.py@@ -545,13 +545,20 @@ def test_v1_4_3_release_blog_is_bilingual_and_verifiable(@pytest.mark.parametrize(('relative', 'href'),(- ('blog/index.html', '/blog/funasr-v1-4-3-pypi-release.html'),- ('en/blog/index.html', '/en/blog/funasr-v1-4-3-pypi-release.html'),+ ('blog/index.html', '/blog/funasr-v1-4-5-pypi-llama-cpp-release.html'),+ ('en/blog/index.html', '/en/blog/funasr-v1-4-5-pypi-llama-cpp-release.html'),),)-def test_blog_index_features_v1_4_3_release(built_site, relative, href):+def test_blog_index_features_latest_release_and_preserves_history(+ built_site, relative, href+):soup = read_soup(built_site / relative)feature = soup.select_one(f'.launch-feature a[href="{href}"]')assert feature- assert 'v1.4.3' in feature.get_text(' ', strip=True)+ assert 'v1.4.5' in feature.get_text(' ', strip=True)+ history = soup.select_one('.previous-release')+ assert history+ history_text = history.get_text(' ', strip=True)+ assert 'v1.4.3' in history_text+ assert 'v1.4.0' in history_text