[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-nvidia-nemotron-3-5-transcribe":3,"topics-all":38,"news-related-039d2494-af49-4d5d-a89f-e3ca91f60df5":67},{"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},"039d2494-af49-4d5d-a89f-e3ca91f60df5","转写没有标准答案:英伟达开源43亿激活语音大模型","英伟达开源语音大模型 Nemotron 3.5 Transcribe:34B 总参激活约 43 亿,40 语言转写、最多 8 人说话人归属,英语榜 4.71% WER 官方称开放权重最佳,单卡每秒转写 2238 秒音频,同步开放 46.3 万小时数据管线 Granary v2,微调单任务最低约半小时。","10 月的语音转写赛道已经挤成红海:Mistral 月底放出 Voxtral 3,IBM 同周更新 Granite Speech 5,再加上 8 月的 Gemini 3.5 Transcribe 和 9 月的 Muse Voice Transcribe,大厂都在往这个口子里塞模型。10 月 28 日,英伟达入场,一次掏出三样东西:开放权重语音大模型 Nemotron 3.5 Transcribe、46.3 万小时的训练数据管线 Granary v2,以及一套微调工具链。\n\n## 「没有唯一正确的转写稿」\n\n这份[技术报告](https:\u002F\u002Fresearch.nvidia.com\u002Flabs\u002Fnemotron\u002FNVIDIA-Nemotron-3.5-Transcribe\u002F)开篇就点破了动机:一段录音不存在唯一正确的转写文本——逐字稿还是书面稿、数字写成「二十五欧元」还是「€25」、要不要保留「呃」「嗯」,取决于谁读、拿来干什么。传统 ASR 在训练时就把这套约定锁死,而它把约定变成了提示词:转写风格、领域词表、说话人标注格式,都靠自然语言指令现场指定,不用重训。\n\n架构上是 SALM 拼法:Transformer 语音编码器,加一个冻结的 Nemotron 3 Diarization 说话人分支(Sortformer,约 1 亿参数、8 个说话人槽位),把「谁在什么时候说话」的活动信息直接注入声学表征,再接 Nemotron 3.5 Lightning 30B-A3B 这个混合 Mamba-Transformer MoE 底座,总参约 340 亿、激活约 43 亿。单模型覆盖 40 种语言转写、31 种语言与英语互译、最多 8 人说话人归属,录音最长 1 小时。\n\n## 跑分很硬,中文会议除外\n\n官方口径:英语 Open ASR Leaderboard 九个公开测试集平均词错率 4.71%,多语言榜 4.62%,自称开放权重最佳;IT之家报道其在英语榜开放权重模型中位列第一。第三方侧,据 IT之家整理:Artificial Analysis 非流式词错率指数 2.8%,与 Voxtral Small Transcribe 并列开源首位;David AI 的 21 语言说话人分离基准上录得 18.26% 的最低错误率。重叠多人说话人场景,官方称据其所知是首个跑赢专用级联系统的 SALM;给定实体词表时,ContextASR-Bench 命名实体错误降六倍。\n\n但同一张表里藏着泼冷水的数字:英文整场会议 cpWER 12.91%,中文会议 AliMeeting 的字符错误率 cpCER 却还有 34.95%。中文多人会议,依然是最难啃的那块骨头。\n\n## 2238 倍实时,还能塞进本地\n\n推理侧,官方基准中 NVFP4 量化版本配 DFlash2 草稿模型,单张 RTX PRO 6000 Blackwell 每秒可转写 2238 秒音频,约 2238 倍实时;模型自带的多 token 预测层就是 vLLM 投机解码的草稿模型。四个精度版本 BF16\u002FFP8\u002FNVFP4\u002FGGUF 全部开放,GGUF 配套 NeMo-Speech.cpp,IT之家报道面向 DGX Spark、Jetson AGX Thor 这类本地设备的版本将陆续跟进。微调也便宜:17 个任务在单张 GB300 上每个只要 0.5-2.3 GPU 小时,13 分钟斯瓦希里语音就能把词错率从 73.7% 拉到 40.8%。\n\n## 转写质量,就是 Agent 成本\n\n附录里最有说服力的是一组语音 Agent 实验:100 个航空客服场景,换上这个模型后任务完成率 58%,对用户语音的词错率 8.6%,对比 Parakeet-CTC-1.1B 的 12.9%;因为听错姓氏导致的工具调用失败从 73 次掉到 2 次,每次交互少 2.7 轮对话、约 3.7 万 token、78 秒。转写每降一个点的错,Agent 就少绕一圈弯路——这笔账比榜单名次直观得多。\n\n边界官方自己也写得清楚:保留底座 LLM 的纯文本能力「明确不是目标」,它是转写专用模型,不是通用助手;PII 脱敏目前还只是合成数据上的小规模探针;34B 总参对真正的端侧也谈不上轻。但「模型 + 46 万小时数据管线 + 微调工具链」这套全家桶,堵的是后来者的路:转写赛道正在从单点模型竞争,变成平台竞争。下一家进场的,不带数据管线恐怕都不好意思开发布会了。","https:\u002F\u002Fresearch.nvidia.com\u002Flabs\u002Fnemotron\u002FNVIDIA-Nemotron-3.5-Transcribe\u002F","474eef8c-e0c3-46cf-adee-c089558220f9",[11,15,18,21],{"id":12,"name":13,"slug":13,"description":14,"color":14},"c4455801-0273-477d-9e92-00b4addfa439","asr",null,{"id":16,"name":17,"slug":17,"description":14,"color":14},"7e89b5cc-57db-4f37-bc6d-28919a73931c","model-release",{"id":19,"name":20,"slug":20,"description":14,"color":14},"8dac812d-3839-4abe-a855-5f56ec9515fd","nvidia",{"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},"74aee232-03b3-4089-9cf5-75226d85ccbd","en","NVIDIA open-sources Nemotron 3.5 Transcribe speech LLM","NVIDIA open-sources Nemotron 3.5 Transcribe: 34B MoE speech LLM, 40 languages, 8-speaker attribution, 4.71% WER, with the Granary v2 data pipeline.","October's speech-transcription lane is crowded: Mistral shipped Voxtral 3 in late October, IBM updated Granite Speech 5 in the same week, and Gemini 3.5 Transcribe and Muse Voice Transcribe landed in August and September. On October 28 NVIDIA entered with three things at once: the open-weights speech LLM Nemotron 3.5 Transcribe, the 463,062-hour training-data pipeline Granary v2, and a fine-tuning toolkit.\n\n## There is no single correct transcript\n\nThe [technical report](https:\u002F\u002Fresearch.nvidia.com\u002Flabs\u002Fnemotron\u002FNVIDIA-Nemotron-3.5-Transcribe\u002F) opens with its premise: a recording has no single correct transcript — verbatim or written form, \"twenty-five euros\" or \"€25\", fillers kept or dropped — it depends on who reads it and for what purpose. Conventional ASR locks that convention in at training time; here the convention becomes a prompt. Transcription style, domain vocabulary, and speaker-tag format are chosen per request in natural language, with no retraining. Granary v2 renders every training recording into at least three transcript variants (verbatim with fillers, written form with fillers, clean written form), so the model learns to follow a requested style instead of guessing one.\n\nArchitecturally it is a SALM: a Transformer speech encoder plus a frozen Nemotron 3 Diarization branch (a Sortformer, about 100M parameters, 8 speaker slots) injects who-spoke-when activity into the acoustic representations, feeding a Nemotron 3.5 Lightning 30B-A3B hybrid Mamba-Transformer MoE backbone — roughly 34B total and 4.3B active parameters. One model covers transcription in 40 languages, translation between English and 31 of them, attribution for up to 8 speakers with annotation in 13 languages, and recordings up to an hour long.\n\n## Strong numbers, except Mandarin meetings\n\nOfficially: a mean WER of 4.71% across the nine public test sets of the English Open ASR Leaderboard and 4.62% on the multilingual board, claimed best among open-weights models; Chinese tech outlet IT Home reports it ranks first among open-weights models on the English board. On the third-party side, per IT Home: 2.8% on Artificial Analysis' non-streaming WER index, tied with Voxtral Small Transcribe at the top of open models, and the lowest 18.26% error on David AI's 21-language speaker-attribution benchmark. For overlapping multi-speaker speech the team claims state-of-the-art accuracy and, to its knowledge, the first SALM to beat specialized cascaded systems; given a list of expected entities, ContextASR-Bench named-entity errors drop six-fold.\n\nThe same results table carries the cold water: full English meetings land at 12.91% cpWER, while Mandarin meetings on AliMeeting still sit at 34.95% cpCER. Multi-speaker Mandarin remains the hardest bone to chew.\n\n## 2,238x real time, and it fits locally\n\nIn NVIDIA's own benchmark, the NVFP4 checkpoint with the DFlash2 draft model transcribes up to 2,238 seconds of audio per second on a single RTX PRO 6000 Blackwell. The model's own multi-token-prediction layer doubles as the speculative-decoding draft in vLLM. All four precision variants — BF16, FP8, NVFP4 and GGUF — are open, with GGUF paired to NeMo-Speech.cpp; IT Home reports builds for local devices such as the DGX Spark and Jetson AGX Thor will follow. Fine-tuning is cheap: 17 tasks each took 0.5-2.3 GPU-hours on a single GB300, and 13 minutes of Swahili audio cut WER from 73.7% to 40.8%. In a medical case study, about 150 hours of synthetic speech plus real accented data brought an Indian-English medical benchmark from 24.03% down to 12.06% — better than Whisper-large-v3 at 14.92% — in roughly 14 GPU-hours.\n\n## Transcription quality is agent cost\n\nThe appendix's most persuasive experiment is a voice-agent run: across 100 airline customer-service scenarios, the agent completed 58% of tasks with this recognizer, at 8.6% WER on user speech versus 12.9% for Parakeet-CTC-1.1B. Tool-call failures caused by misrecognized surnames fell from 73 to 2, and each interaction saved 2.7 turns, about 37,000 LLM tokens, and 78 seconds. Every point of WER you shave is a detour the agent no longer takes — a more concrete account than any leaderboard rank.\n\nThe limits are stated plainly: preserving the base LLM's text-only abilities is \"explicitly not a goal\" — this is a transcription specialist, not a general assistant; PII redaction is still a small, synthetic, English-only probe; and 34B total parameters is not exactly light for the edge. But the bundle — a model plus a 463k-hour data pipeline plus tooling — is a platform play: transcription is shifting from single-model races to stack wars. The next entrant without a data pipeline may not even bother holding a launch event.","nvidia-nemotron-3-5-transcribe","2026-10-29T15:15:27Z","2026-10-29T15:15:38.310548Z","2026-10-29T15:15:38.310558Z",true,"agent",318,[39,49,58],{"slug":40,"tag_slug":41,"title_zh":42,"title_en":43,"intro_zh":44,"intro_en":45,"id":46,"is_active":35,"created_at":47,"modified_at":48},"image-generation","text-to-image","图像生成：从文生图到逐框编辑的模型竞赛","Image Generation: from text-to-image to frame-level editing","文生图已是红海，战场正转向编辑与可控性：FLUX.2\u002F3 用开源和量化把生图搬进消费级显卡，Qwen-Image、Seedream 等国产模型紧咬不放，GPT Image 2 则把推理能力带进像素。本专题追踪图像生成与编辑模型的发布、架构演进、加速量化与商业化数据，看「好图」如何变成「可用的图」。","Text-to-image is a crowded race, and the frontier has moved to editing and control: FLUX.2\u002F3 push open weights and quantization onto consumer GPUs, Qwen-Image and Seedream close in from China, and GPT Image 2 brings reasoning into pixels. This topic tracks model releases, architecture shifts, acceleration work, and the business numbers behind \"pretty\" becoming \"usable\".","52de7c59-7864-4569-8ae7-3d656d26a1a1","2026-11-20T07:20:23.534506Z","2026-11-20T07:20:23.534527Z",{"slug":50,"tag_slug":50,"title_zh":51,"title_en":52,"intro_zh":53,"intro_en":54,"id":55,"is_active":35,"created_at":56,"modified_at":57},"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":59,"tag_slug":59,"title_zh":60,"title_en":61,"intro_zh":62,"intro_en":63,"id":64,"is_active":35,"created_at":65,"modified_at":66},"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":68},[69,74,79,84,89,94],{"id":70,"title":71,"news_slug":72,"published_at":73},"503fc4f0-755d-43f1-89cc-8d86b4cdaed9","7.6M参数土耳其语TTS:WER 1.09%却没人做过盲听","antalia-2-mini-turkish-tts-synthetic-data","2026-11-12T01:40:00+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"34bd9f83-04c2-4ffa-8e68-f65a74dfb490","开源语音智能体破50分:Apriel Voice 27B","servicenow-apriel-voice-27b-preview","2026-11-02T13:19:00+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"265c8b9e-9a65-4198-bc00-a6ba1c8b8248","英伟达开源 4GB 小模型:Bolt E4B 少说话多干活","nemotron-bolt-e4b-4gb-edge-slm","2026-10-24T21:09:01+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"35e9f682-7952-4f6a-a8e9-7f8bdd47e791","IBM 语音三件套更新:2B 旗舰 ASR 精度开源第二","granite-speech-5-collection-update","2026-10-22T21:08:48+00:00",{"id":90,"title":91,"news_slug":92,"published_at":93},"d83c4910-6dce-4171-8a44-9f68c9f841af","Qwen-Audio-3.1-ASR开源:转写自带说话人","qwen-audio-3-1-asr-speaker-diarization","2026-10-22T17:13:29+00:00",{"id":95,"title":96,"news_slug":97,"published_at":98},"2ddb354a-06f7-4781-b2d5-bcfd1faf38a5","Dory 1.0:英伟达4B循环模型,指令遵循反超9B","nvidia-dory-1-0-4b-looped-llm","2026-10-20T21:09:38+00:00"]