[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-granite-speech-5-turbo-ctc-470m":3,"topics-all":38,"news-related-454286bd-cb8e-462e-8b33-1b4c77b27262":57},{"id":4,"title":5,"summary":6,"content":7,"original_url":8,"source_id":9,"tags":10,"translations":24,"news_slug":31,"published_at":32,"created_at":33,"modified_at":34,"is_published":35,"publish_type":36,"image_url":14,"view_count":37},"454286bd-cb8e-462e-8b33-1b4c77b27262","470M 语音模型 1 秒转写 3.5 小时:IBM 把 ASR 里的语言模型砍掉了","IBM 开源 Granite Speech 5.0 Turbo CTC:470M 参数 encoder-only 英文 ASR,H200 上吞吐超 12,600 RTFx,批量推理 1 秒转写 3.5 小时以上语音;OpenASR 公共榜 WER 4.85%\u002F5.00%,比前代快 20 倍以上。","一台 H200 GPU,一秒钟,转写完 3.5 个小时以上的英文语音。这不是推理集群的聚合吞吐,而是单个 470M 参数模型——IBM 刚在 Hugging Face 开源的 Granite Speech 5.0 Turbo CTC——用批量推理跑出来的官方数字。在所有人都在往语音模型里塞 LLM 的 2026 年,IBM 的做法恰好相反:把语言模型整个砍掉。([官方发布博客](https:\u002F\u002Fhuggingface.co\u002Fblog\u002Fibm-granite\u002Fgranite-speech-5-0-470m-turboctc))\n\n## 速度与准确率,这次真的都要\n\n官方口径:在 OpenASR Leaderboard 的公共英文短格式测试集上,非商用版(聚合 WER 4.85%)和 Apache 2.0 版(5.00%)两个变体,聚合吞吐都超过 12,600 RTFx。作为对照,官方博客明确写了\"比之前的 Granite Speech 模型快 20 倍以上\"。远场场景的 FFASR 榜单上,两个变体分别排第 5 和第 9,同时也是该榜上最快的两个模型(截至 8 月 25 日)。\n\n需要说明:这些是 IBM 自己提交的官方结果(official results),不是第三方复测,读数时请保留这一层归因。\n\n## 砍掉语言模型,是一次反向操作\n\n上一代 Granite Speech 的架构是三段式:声学编码器 + 投影层 + 带 LoRA 适配器的 Granite 语言模型。这一代直接砍成 encoder-only——16 个 Conformer 块堆叠,第 8 块输出做 self-conditioning,注意力用 chunkwise 方式避免随序列长度二次增长,训练目标是经典的 CTC loss。\n\n代价也写得明明白白:LM 版独有的语音翻译、关键词偏置能力没有了。换来的东西很实在:470M 的内存脚印,加上适合边缘设备的部署画像。官方还配了一个在浏览器里跑流式识别的 WebGPU demo(仅支持 Chrome\u002FEdge)。\n\n## 低 token rate 才是提速的钥匙\n\n架构上最有意思的一处改动在 token 生成率:前代编码器每秒输出 50 个字符,Granite 5.0 降到每秒 12.5 个 token。从 100fps 的 log Mel 频谱前端降到 12.5 token\u002Fs,靠的是三级 2 倍下采样:第一级把相邻特征向量 reshape 堆叠,第二、三级内嵌在前两个 Conformer 块里,用 stride=2 的时间卷积完成。输出 tokenizer 也分了家:非商用版用 SentencePiece,Apache 版用 BPE。\n\n序列长度直接决定注意力与解码的开销。token 率砍到四分之一,再叠加 chunkwise 注意力,20 倍的吞吐提升就不是魔法,而是算术。\n\n## 数据配方:开源语料打底,合成数据补角\n\n训练数据一侧,两个变体共享约 5.8 万小时自然语料:MLS(44,600 小时)、YODAS(8,900 小时)、CommonVoice-17(2,500 小时)、Librispeech(960 小时)、VoxPopuli(500 小时)、AMI(150 小时)、Earnings-22(100 小时)。非商用版额外加了 GigaSpeech(10,000 小时)和 SPGI Speech(4,900 小时)——这也是它 WER 略低的直接原因。\n\n合成数据补了三个角:2,000 小时多说话人拼接语料(来自 MLS\u002FYODAS\u002FCommonVoice\u002FVoxPopuli\u002FAMI)、500 小时会议场景拼接(Earnings-22),以及 240 小时专门针对数字、货币、网址、电话号码、地址这类 ASR 传统弱项的内容——文本用 gpt-oss-120b 或 gpt-oss-20b 生成,再用 StyleTTS2 合成为语音。用开源 LLM 制造 ASR 训练数据,这个闭环本身值得玩味。\n\n## 所以呢\n\n过去两年语音识别的主流叙事是\"越大越好\":编码器后面挂越来越大的语言模型,用 LLM 的世界知识兜底。Granite Speech 5.0 Turbo CTC 给出了反方向的证据:如果你的场景就是转写本身,一个 470M、encoder-only、低 token 率的模型,可以在吞吐上拉开数量级差距,还把商用友好的 Apache 2.0 许可留给你。\n\n两个变体的取舍也摆得清楚:要极致准确率就拿非商用版的数据优势(许可为 CC-BY-NC-SA,不能商用),要落地就选 Apache 2.0 版,多付出 0.15 个百分点的 WER。对做边缘设备、实时字幕、批量转档的团队,这是一次可以直接抄的工程示范:先把 token 率降下来,再谈别的。","https:\u002F\u002Fhuggingface.co\u002Fblog\u002Fibm-granite\u002Fgranite-speech-5-0-470m-turboctc","653dda08-2edc-4d17-aeb2-56b0c88dd918",[11,15,18,21],{"id":12,"name":13,"slug":13,"description":14,"color":14},"40269b40-7942-4650-9672-ed2e6524d37a","ai-technology",null,{"id":16,"name":17,"slug":17,"description":14,"color":14},"0ef8513a-0a26-42f0-b6f9-5b6dadded45c","efficiency",{"id":19,"name":20,"slug":20,"description":14,"color":14},"7e89b5cc-57db-4f37-bc6d-28919a73931c","model-release",{"id":22,"name":23,"slug":23,"description":14,"color":14},"b9bd9039-fcdb-41a8-b85b-fc1587def2b9","open-source",[25],{"id":26,"lang":27,"title":28,"summary":29,"content":30},"79720a43-848a-4aca-b8d9-55182211aef6","en","470M Speech Model Transcribes 3.5 Hours in One Second: IBM Cuts the LM Out of ASR","IBM open-sources Granite Speech 5.0 Turbo CTC: a 470M encoder-only English ASR model reaching over 12,600 RTFx on an H200 — transcribing 3.5+ hours of speech in one second of batched inference, with 4.85%\u002F5.00% WER variants.","One H200 GPU, one second, and more than 3.5 hours of English speech transcribed. That is not aggregate cluster throughput — it is the official number from a single 470M-parameter model, IBM's newly open-sourced Granite Speech 5.0 Turbo CTC on Hugging Face, running batched inference. In a year when everyone keeps bolting LLMs onto speech models, IBM went the opposite way: it removed the language model entirely. ([Official release blog](https:\u002F\u002Fhuggingface.co\u002Fblog\u002Fibm-granite\u002Fgranite-speech-5-0-470m-turboctc))\n\n## Speed and accuracy, both this time\n\nPer the official figures on the public English short-form test sets of the OpenASR Leaderboard, the two variants — the non-commercial one (4.85% aggregate WER) and the Apache 2.0 one (5.00%) — both deliver aggregate throughput in excess of 12,600 RTFx. For context, the blog states this is over 20x faster than previous Granite Speech models. On the far-field FFASR leaderboard, the two variants rank 5th and 9th in accuracy as of August 25 while also being the two fastest models on that board.\n\nOne caveat worth keeping: these are IBM's own official submissions, not third-party reproductions.\n\n## Removing the LM is the actual story\n\nThe previous Granite Speech architecture was a three-stage pipeline: an acoustic encoder, a projector, and a Granite language model with LoRA adapters. This generation collapses to encoder-only — a stack of 16 Conformer blocks, self-conditioning at the output of the 8th block, chunkwise attention to avoid quadratic scaling with sequence length, and classic CTC loss as the training objective.\n\nThe trade-off is stated plainly: capabilities unique to the LM-equipped models, such as speech translation and keyword biasing, are gone. What you get instead is a 470M memory footprint and a deployment profile suited to edge devices. IBM also ships a WebGPU demo of streaming recognition that runs in the browser (Chrome or Edge only).\n\n## Low token rate is the real key\n\nThe most interesting architectural change is the output token rate: previous Granite encoders emitted 50 characters per second, while Granite 5.0 generates 12.5 tokens per second. Getting from a 100fps log-Mel spectrogram front end down to 12.5 tokens\u002Fs relies on three stages of 2x subsampling — the first stacks adjacent feature vectors via a reshape, while the second and third are built into the first two Conformer blocks using strided temporal convolutions. The tokenizers also split: the non-commercial model uses SentencePiece, the Apache 2.0 model uses BPE.\n\nSequence length drives attention and decoding costs. Cut the token rate to a quarter, stack chunkwise attention on top, and a 20x throughput jump stops being magic and becomes arithmetic.\n\n## The data recipe\n\nBoth variants share roughly 58,000 hours of natural corpora: MLS (44,600 hours), YODAS (8,900), CommonVoice-17 (2,500), Librispeech (960), VoxPopuli (500), AMI (150), and Earnings-22 (100). The non-commercial variant adds GigaSpeech (10,000 hours) and SPGI Speech (4,900) — which is exactly why its WER is slightly lower.\n\nSynthetic data fills three gaps: 2,000 hours of multi-speaker concatenations from MLS\u002FYODAS\u002FCommonVoice\u002FVoxPopuli\u002FAMI, 500 hours of meeting-style concatenations from Earnings-22, and 240 hours targeting traditional ASR weak spots — numbers, currencies, website names, phone numbers, and addresses — with text generated by gpt-oss-120b or gpt-oss-20b and synthesized into speech via StyleTTS2. Using open-weight LLMs to manufacture ASR training data is a loop worth pondering on its own.\n\n## So what\n\nThe dominant speech-recognition narrative of the past two years has been \"bigger is better\": hang an ever-larger language model behind the encoder and let LLM world knowledge do the patching. Granite Speech 5.0 Turbo CTC offers evidence in the other direction: if your use case is transcription itself, a 470M encoder-only model with a low token rate can open up an order-of-magnitude throughput lead while keeping a commercially friendly Apache 2.0 license.\n\nThe variant choice is laid out cleanly: take the non-commercial build for maximum accuracy (CC-BY-NC-SA, no commercial use), or take Apache 2.0 for deployment and pay 0.15 WER points extra. For teams building edge devices, live captions, or bulk transcription pipelines, this is an engineering lesson you can copy directly: lower the token rate first, then talk about everything else.","granite-speech-5-turbo-ctc-470m","2026-08-31T15:10:00Z","2026-08-31T15:08:27.096840Z","2026-08-31T15:08:27.096848Z",true,"agent",164,[39,48],{"slug":40,"tag_slug":40,"title_zh":41,"title_en":42,"intro_zh":43,"intro_en":44,"id":45,"is_active":35,"created_at":46,"modified_at":47},"ai-for-science","AI for Science 2026：从 UniPert 到 GPT-Rosalind 的硬核进化","AI for Science 2026: from UniPert to GPT-Rosalind","生命科学、化学材料、物理世界模型——AI 正在从\"语言工具\"变成\"实验伙伴\"。本专题收录 AI 在三大科学方向的关键节点：UniPert 统一基因与化学扰动空间、GPT-Rosalind 端到端生命科学推理、达摩院 AI 智能体 28 小时找到 4 种超导新材料、Anthropic Claude Science 把工作台做成标准品。","From language tool to lab partner — AI is reshaping life sciences, chemistry\u002Fmaterials, and physical world models. This topic covers the key milestones: UniPert unifying genetic-chemical perturbation spaces, GPT-Rosalind's end-to-end life-sciences reasoning, DAMO's AI agent discovering 4 superconducting materials in 28 hours, and Anthropic's Claude Science workbench going mainstream.","988a4300-5fab-41c4-b5d8-63711a2dc757","2026-09-10T01:34:15.296649Z","2026-09-10T01:34:15.296663Z",{"slug":49,"tag_slug":49,"title_zh":50,"title_en":51,"intro_zh":52,"intro_en":53,"id":54,"is_active":35,"created_at":55,"modified_at":56},"h3-series","MiniMax H3 系列：从开源权重到 35 倍吞吐","MiniMax H3 Series: from open weights to 35x throughput","MiniMax H3 自 2026 年 8 月开源以来节奏密集：官方把生成、参考与编辑收回一个模型；ComfyUI 当天压进 RTX 3060；摩尔线程 3 小时完成国产 GPU 适配；fal 后训练版把吞吐拉到 35 倍；FastH3 蒸馏再砍推理成本。本专题持续追踪 H3 的发布—开源—蒸馏—部署全链路。","Since MiniMax open-sourced H3 in August 2026 the pace has been relentless: one unified omni-modal model, same-day ComfyUI support down to an RTX 3060, a 3-hour Day-0 port to Moore Threads GPUs, fal's post-trained H3 Max at 35x throughput, and FastH3 distillation cutting inference cost further. This topic tracks the full H3 chain — release, open weights, distillation, deployment.","83ef0daa-3c31-4cb3-86ed-e5ee58654d5f","2026-09-08T07:33:19.942193Z","2026-09-08T07:33:19.942209Z",{"items":58},[59,64,69,74,79,84],{"id":60,"title":61,"news_slug":62,"published_at":63},"30fca629-bace-4832-9789-b44aa8c8989d","学生团队从零训出开源 7B 模型 ZGCM-1:数学推理硬刚 235B 前沿","zgcm-1-open-7b-foundation-model","2026-09-15T19:10:00+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"108af093-b226-4372-9cf0-77323ffc5456","小鹏 X-AuT 给语音大模型剪枝:音频塔砍 4 层,车载推理提速 21.4%","xpeng-x-aut-audio-encoder-pruning","2026-09-12T19:06:47+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"70ea74fc-77ed-49a2-8498-f24edd822970","AI 工具把 Linux 内核挖出 2000 个 CVE,维护者快扛不住了","linux-kernel-ai-cve-overflow","2026-09-09T05:00:00+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"2b37a19b-1dde-4238-bef5-39b1d19157f1","OpenBMB 开源 MiniCPM5-2B:2B 端侧模型平均分超对比集 4B 级","openbmb-minicpm5-2b-on-device","2026-09-07T17:02:00+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"58ed753e-ad6d-4aac-95f4-36bf217e169c","把 10 万条人类视频变成机器人教材:RoboTok 检索 mAP 提升约 50 倍,hard 任务 79.3% 对 19.5%","robotok-retrieval-benchmark-reread","2026-09-06T21:11:25+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"dc2f4ead-963c-4a8e-bd41-400bebf83bb4","物理、几何、外观一个模型全包:Puffin-World 开源,相机 roll 误差低至 0.26°","puffin-world-native-3d-world-states","2026-09-06T19:09:41+00:00"]