[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-qwen-fun-asr-realtime":3,"topics-all":36,"news-related-d67a8083-5341-437a-a853-2057dfb8416b":55},{"id":4,"title":5,"summary":6,"content":6,"original_url":7,"source_id":8,"tags":9,"translations":23,"news_slug":29,"published_at":30,"created_at":31,"modified_at":32,"is_published":33,"publish_type":34,"image_url":13,"view_count":35},"d67a8083-5341-437a-a853-2057dfb8416b","通义 Fun-ASR-Realtime：首字延迟压进百毫秒，100 小时直播实战验证","通义实验室 7 月 6 日升级了端到端语音识别大模型的实时版——Fun-ASR-Realtime。流式识别最大的技术难题是「首字延迟」与「准确率」的此消彼长：流式解码天然有延迟，但流得越快往往越容易出错。Fun-ASR-Realtime 给出的答案是百毫秒级首字延迟 + 接近离线模型的识别准确率。它走的是端到端架构 + RAG 检索增强的组合，原生支持大规模热词自定义、敏感词\u002F语气词过滤、ITN 归一化和智能标点预测，让流式输出从第一时刻起就接近「可直接发布」的文本。语言覆盖上一次性支持 16 种方言和 30 种语言，包括中英文及东亚、东南亚、南亚、中东、欧洲主流语种，中英混说场景无需预设语种标签。\n\n更具说服力的是 7 月 2 日结束的影视飓风「重返荒岛」100 小时直播：Fun-ASR-Realtime 全程提供实时字幕，生成 6 万多条字幕、累计 132 万字，几乎没有出现延迟堆积或语义断裂——这相当于一次工业级压力测试。\n\n把视野放大，流式语音识别已是 2026 年各家大模型的标配战场，OpenAI GPT-Realtime、Anthropic Claude Voice、Google Gemini Live Translate 都在压延迟。通义选择把「流式 + 方言 + 多语言」一次打通，差异化明显。API 已在千问云开放，定价 ¥0.00033\u002F秒，按 100 小时直播体量估算单场成本约百元，已是工程化部署可承受的范围。","https:\u002F\u002Fwww.qianwenai.com\u002Fmodels\u002Ffun-asr-realtime","c36a21ac-2a77-421b-9519-1e150695732a",[10,14,17,20],{"id":11,"name":12,"slug":12,"description":13,"color":13},"0ef8513a-0a26-42f0-b6f9-5b6dadded45c","efficiency",null,{"id":15,"name":16,"slug":16,"description":13,"color":13},"499f4b56-819d-49a3-9609-33e775143b86","multimodal",{"id":18,"name":19,"slug":19,"description":13,"color":13},"b1853a5a-d940-42b7-94f9-0488ee3f2cf7","new-model",{"id":21,"name":22,"slug":22,"description":13,"color":13},"c187600e-804c-4697-b828-1e4330e0eb10","qwen",[24],{"id":25,"lang":26,"title":27,"summary":28,"content":13},"c4922a62-fb08-4068-884e-88fca79c3280","en","Fun-ASR-Realtime: sub-100ms first word, 100 live hours tested","Tongyi Lab on July 6 upgraded the real-time version of its end-to-end speech recognition LLM — Fun-ASR-Realtime. The biggest technical challenge of streaming recognition is the trade-off between \"first-character latency\" and \"accuracy\": streaming decoding is inherently delayed, but the faster you stream, the more error-prone you tend to be. Fun-ASR-Realtime's answer is sub-100ms first-character latency + accuracy close to offline models. It takes the combination of end-to-end architecture + RAG retrieval enhancement, natively supporting large-scale hotword customization, sensitive-word\u002Ffiller-word filtering, ITN normalization, and intelligent punctuation prediction, making the streaming output from the first moment close to \"directly publishable\" text. Language coverage supports 16 dialects and 30 languages in one go, including Chinese and English as well as major East Asian, Southeast Asian, South Asian, Middle Eastern, and European languages, with no need to preset language tags for Chinese-English mixed scenes. Even more convincing is the 100-hour \"Return to the Deserted Island\" live broadcast that ended on July 2: Fun-ASR-Realtime provided real-time subtitles throughout, generating more than 60,000 subtitles, accumulating 1.32 million characters, with almost no latency pileup or semantic breaks — equivalent to an industrial-grade stress test. Zooming out, streaming speech recognition is now the standard battlefield for the major model vendors in 2026, with OpenAI GPT-Realtime, Anthropic Claude Voice, Google Gemini Live Translate all squeezing latency. Tongyi's choice to crack \"streaming + dialects + multilingual\" in one shot has clear differentiation. The API is already open on Qianwen Cloud, priced at ¥0.00033\u002Fsecond, with a single 100-hour live broadcast estimated to cost about 100 yuan, already within the affordable range for engineering deployment.","qwen-fun-asr-realtime","2026-07-06T15:00:00Z","2026-07-06T16:06:57.935091Z","2026-08-19T02:08:40.142862Z",true,"agent",234,[37,46],{"slug":38,"tag_slug":38,"title_zh":39,"title_en":40,"intro_zh":41,"intro_en":42,"id":43,"is_active":33,"created_at":44,"modified_at":45},"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":47,"tag_slug":47,"title_zh":48,"title_en":49,"intro_zh":50,"intro_en":51,"id":52,"is_active":33,"created_at":53,"modified_at":54},"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":56},[57,62,67,72,77,82],{"id":58,"title":59,"news_slug":60,"published_at":61},"dff241e0-7bfa-499b-b811-6869690b196d","Qwen3.5-LiveTranslate-Flash：阿里同传模型升级，延迟降至2.8秒，支持60语言与实时语音克隆","qwen3-5-livetranslate-flash-2-8s-60-lang","2026-05-24T04:10:00+00:00",{"id":63,"title":64,"news_slug":65,"published_at":66},"604c331d-9627-4f95-a721-80e83d0b5ba7","Meta 发布 Muse Spark：原生多模态推理模型革新视觉理解","meta-muse-spark-msl-native-multimodal-screenspot-72","2026-04-26T16:05:00+00:00",{"id":68,"title":69,"news_slug":70,"published_at":71},"7fae0753-96df-4d4b-9ebd-cf0509c08b37","LLM架构演进：从规模竞赛到效率优化的范式转变","llm-architecture-evolution-2026-moe-multimodal-turboquant","2026-04-25T04:12:00+00:00",{"id":73,"title":74,"news_slug":75,"published_at":76},"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":78,"title":79,"news_slug":80,"published_at":81},"3a419e1c-fb2b-467e-a5b6-a7188525bc03","TraceForge:推理痕迹自己造,学生反超教师真迹","traceforge-family-native-distillation","2026-10-22T15:10:42+00:00",{"id":83,"title":84,"news_slug":85,"published_at":86},"15da3422-7db5-4da6-a12a-e4fd84b9a50b","视觉脚手架:88M 参数把视觉特征搬进思维链","visual-scaffolding-native-multimodal-reasoning","2026-10-20T23:03:57+00:00"]