[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-tencent-hunyuan-hy-asr-3-0-preview-moe-speech-lm":3,"topics-all":38,"news-related-64b79fb5-b416-4b58-8402-0bc9ad22a27d":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},"64b79fb5-b416-4b58-8402-0bc9ad22a27d","腾讯混元 Hy ASR 3.0 preview：MoE + 语音 Encoder 与 LLM 联合训练，把 ASR 推进到'理解语境'","腾讯混元 8 月 4 日发布语音识别模型 Hy ASR 3.0 preview,基于新一代大语言模型 Hy3,在多语种开源评测中 WER 控制在 3% 左右(普通话 3.34%、英语 2.62%、粤语 3.12%)。架构上采用 MoE 并自研无监督语音 Encoder,通过数千万小时级联合训练把语音识别从'逐字转写'推进到'理解语境、场景直出'。","## 技术背景\n\n语音识别过去几年最大的瓶颈不是'识别准不准',而是'懂不懂用户在说什么'。传统 ASR 走的是逐字转写 + 后处理纠错的路,遇到同音词、上下文断句、专业术语、噪声音学场景时几乎只能依赖规则匹配——一旦用户说长句、口语、夹杂方言,模型就只能听之任之。\n\nGPT 类大语言模型兴起后,'ASR + LLM 后处理'成为行业共识:把识别结果丢给 LLM 做纠错。但这条路线有天然缺陷——LLM 拿不到音频里的副语言信息(语气、停顿、背景声),纠错只能在文本层面打补丁。\n\n腾讯混元 Hy ASR 3.0 preview 选的是第三条路:**ASR Encoder 与 LLM 联合训练**,把'听懂'前置到声学表征层面。8 月 4 日正式发布,多家媒体(IT 之家、36 氪、量子位、网易、凤凰科技)同步报道。\n\n## 核心能力\n\nHy ASR 3.0 preview 在开源评测集上把多语种 WER(词错误率)都打到 3% 量级:中文普通话 **3.34%**、英语 **2.62%**、粤语 **3.12%**——这一组数字在多家独立来源的报道里完全一致,不是单家宣传口径(交叉来源:量子位官方报道、IT 之家、36 氪新闻快讯、腾讯新闻)。自建评测集在通用识别、方言识别、上下文理解、高噪 \u002F 耳语复杂声学场景下,WER 也保持较低水平,'整体领先竞品'(同源措辞)。\n\n四项重点能力提升:**通用识别更准确**(减少错字漏字)、**更理解用户意图**(上下文纠错同音词、消除语义歧义)、**专业场景适配**(支持热词注入,识别品牌名 \u002F 人名 \u002F 行业术语,降低业务接入和持续维护成本)、**复杂环境稳定**(高噪、耳语、悄悄话场景下专项优化)。\n\n## 架构与训练\n\n能力提升不是单模块变化,而是架构、数据、后训练三层叠加的结果。\n\n**架构层**——模型采用 **MoE 架构**(兼顾效率与性能),基座模型从上一代 Hy2 升级到最新一代 **Hy3**,进一步增强语言理解、上下文建模和语义推理能力。语音侧,团队**自研无监督语音 Encoder**,通过数千万小时级无监督语音数据训练,使其能从复杂音频中提取高质量的声学表征。\n\n**数据层**——对语音 Encoder 与大语言模型进行**联合训练**,引入数千万小时级、多来源的语音数据,覆盖多种方言、口音和声学环境,通过高质量数据管线进行精细化标注。在方言维度,SFT 数据体系进一步覆盖 **10 大方言片区 + 20 余个二级小片区**。\n\n**后训练层**——围绕通用识别、上下文理解和复杂场景鲁棒性,团队构建了高质量 SFT recipe;并通过**多阶段强化学习(RL)**,分别针对通用转写准确性、Any-context 上下文能力、复杂长尾场景做优化,降低模型在复杂声学环境中的误识别与漏识别。\n\n## 落地与商业\n\nHy ASR 3.0 preview 已在腾讯云官网对外提供 API 服务( 参数 ),广泛应用于智能客服、内容理解、语音搜索等场景。\n\n产品侧节奏:**腾讯元宝(yuanbao)首发上线**并以免费形式对外开放,用户打开元宝按住说话即可体验方言识别、上下文智能纠错与复杂环境稳定转写;WorkBuddy 等产品也在陆续接入中。\n\n## 评论\n\n腾讯把 ASR 与 LLM **联合训练而不是松耦合串联**,是这次工程上最值得关注的决定。Whisper 类传统方案是'ASR 转写 → LLM 后处理'两步走,文本层面纠错终究是事后打补丁;而 Hy ASR 3.0 preview 在声学表征阶段就携带语言知识——这意味着方言、口音、噪声环境下的转写不再是'先错后改',而是直接从表征空间规避错误。\n\n对行业来说,3% 量级的 WER 已经接近人类速记员水平(中文普通话人工速记 WER 通常在 2-4% 之间)。**问题不再是模型识别不识别得出来,而是模型能多快进入具体场景**——客服、金融、医疗、法律这种术语密集的业务,谁先把'热词注入'做成开箱即用、按需调用的产品,谁就能吃掉 ASR 增量市场。\n\n回到产品节奏——这一年内腾讯混元密集发布 **Hy3 大语言基座**、**Hy ASR 3.0**、**AngelSpec 投机解码框架**、**Marvis 端侧 Agent**,'底层基座 + 中层 ASR\u002F翻译 + 上层 Agent 应用'的全栈布局已经成形。当大模型竞争从'谁跑分高'切换到'谁先吃掉具体业务'时,全栈协同的厂商反而更容易拿到下一个周期的入场券。\n\n### 引用源\n\n- 量子位原文(主要信源):https:\u002F\u002Fwww.qbitai.com\u002F2026\u002F08\u002F465973.html\n- 36 氪新闻快讯:https:\u002F\u002F36kr.com\u002Fnewsflashes\u002F3924966551861640\n- IT 之家:https:\u002F\u002Fwww.ithome.com\u002F0\u002F985\u002F538.htm\n- 凤凰科技:https:\u002F\u002Ftech.ifeng.com\u002Fc\u002F8vJlqs7btgR\n- 网易科技:https:\u002F\u002Fwww.163.com\u002Ftech\u002Farticle\u002FL3GNI8MB00098IEO.html","https:\u002F\u002Fwww.qbitai.com\u002F2026\u002F08\u002F465973.html","3bd971a8-3897-43d9-84ac-43879efd2f94",[11,15,18,21],{"id":12,"name":13,"slug":13,"description":14,"color":14},"fca9258a-9430-455a-b95d-b9fae5e373a8","ai-inference",null,{"id":16,"name":17,"slug":17,"description":14,"color":14},"5e628969-6d2a-437f-998a-104e4b16cfb1","ai-progress",{"id":19,"name":20,"slug":20,"description":14,"color":14},"01598627-1ea6-4b27-a5d8-874971571a71","llm",{"id":22,"name":23,"slug":23,"description":14,"color":14},"7e89b5cc-57db-4f37-bc6d-28919a73931c","model-release",[25],{"id":26,"lang":27,"title":28,"summary":29,"content":30},"b05c6597-8e90-4325-93ae-91eeabe918c8","en","Tencent Hy ASR 3.0: trained with the LLM, ASR learns context","Tencent Hunyuan released Hy ASR 3.0 preview on Aug 4. The model is built on the latest Hy3 LLM, hits a WER of about 3 percent across languages in open-source evaluations (Mandarin 3.34%, English 2.62%, Cantonese 3.12%). It uses a Mixture-of-Experts backbone with a self-developed unsupervised speech Encoder, and the encoder and LLM are jointly trained on tens of millions of hours of speech data.","## Background\n\nFor the past few years, the real ceiling on automatic speech recognition has not been whether the model can transcribe accurately; it is whether it can understand what the user actually said. Classical ASR systems rely on letter-by-letter transcription plus rule-based post-processing. They collapse the moment users produce long utterances, casual speech, code-mixed sentences, or content packed with homophones, professional terms, and noisy acoustics.\n\nAfter GPT-style large language models became practical, the dominant pattern was a two-stage pipeline: ASR first, LLM post-correction next. That buys something, but it is structurally limited; the LLM never sees the audio, so it can only patch what the ASR has already mangled.\n\nTencent Hunyuan Hy ASR 3.0 preview, officially announced on August 4, takes a third path. It trains the **speech Encoder jointly with the LLM**, moving comprehension up into the acoustic representation layer. Coverage includes IT Home, 36Kr, QbitAI, NetEase, and iFeng.\n\n## Headline numbers\n\nOn multiple open-source evaluation sets, Hy ASR 3.0 preview drives Word Error Rate down to roughly 3% across languages: **3.34% Mandarin**, **2.62% English**, **3.12% Cantonese**. The same figures appear across independent reports, not just vendor talking points. On Tencent's internal evaluation suite covering general recognition, dialect recognition, context understanding, and high-noise \u002F whispered acoustics, WER remains low and is described as overall best-in-class relative to comparable commercial systems.\n\nFour capability axes are emphasized: more accurate general recognition (fewer wrong or missing characters), better intent understanding (context-aware correction of homophones and semantic disambiguation), smoother adaptation to professional scenarios (hotword injection for brand names, person names, industry terms, reducing business-side integration and maintenance cost), and stable behaviour under acoustically hard conditions (high noise, whispering, muttering).\n\n## Architecture and training\n\nThe lift comes from three layers compounding on top of each other rather than a single module swap.\n\n**Architecture.** The backbone is a **Mixture-of-Experts** design that balances efficiency and capacity. The base LLM moves from the previous generation to the latest **Hy3**, strengthening language understanding, contextual modelling and semantic reasoning. On the speech side, the team built a **self-developed unsupervised speech Encoder** that learns from tens of millions of hours of unlabelled speech to pull high-quality acoustic representations out of complex audio.\n\n**Data.** The Encoder and the LLM are jointly trained on tens of millions of hours of multi-source speech data, spanning a wide set of dialects, accents and acoustic environments, with careful annotation through a high-quality pipeline. For dialect coverage, the SFT data system reaches **10 major dialect regions** plus **20+ second-level sub-regions**.\n\n**Post-training.** A high-quality SFT recipe is constructed around general recognition, context understanding and acoustic robustness. The team further applies **multi-stage reinforcement learning**, with separate objectives for general transcription accuracy, any-context understanding, and complex long-tail acoustic scenarios, targeting the residual mis-recognition and missed-recognition cases.\n\n## Deployment\n\nHy ASR 3.0 preview is already available via Tencent Cloud as an API service (engine_model_type parameter ), targeting intelligent customer service, content understanding, voice search and similar use cases.\n\nOn the product side, **Tencent Yuanbao** is the first consumer surface to integrate the model and is offering it free during the launch window. Users can hold to talk and experience dialect recognition, context-aware correction, and stable behaviour in noisy environments. WorkBuddy and other products are integrating in waves.\n\n## Why it matters\n\nThe most consequential engineering decision in this release is that Tencent chose to **jointly train** the ASR Encoder and the LLM rather than splice them together. With Whisper-style pipelines, ASR transcribes first and the LLM patches the text afterwards; the LLM never has access to acoustic cues, so it can only patch what the ASR has already produced. Hy ASR 3.0 preview, by contrast, bakes language knowledge into the acoustic representation itself. The benefit shows up most clearly in dialects, accents and noisy conditions: errors get avoided at the embedding stage rather than corrected in text afterwards.\n\nAt roughly 3% WER, the model is in the range of experienced human stenographers for Mandarin (typically 2-4%). The next competitive question is therefore not who transcribes more accurately but who can reach specific business contexts fastest. Whoever ships **hotword injection as a turn-key, on-demand product** for call centres, finance, healthcare, and law will eat most of the ASR incremental market.\n\nLooking at Tencent Hunyuan's release cadence over the past year: **Hy3 base LLM**, **Hy ASR 3.0**, **AngelSpec speculative decoding framework**, **Marvis on-device agent**; the company is now running a full-stack play of base models, mid-layer speech and translation systems, and an upper-layer agent application surface. Once the LLM race shifts from leaderboard numbers to vertical landing, full-stack vendors are the ones best positioned for the next cycle.\n\n### Sources\n\n- QbitAI primary article: https:\u002F\u002Fwww.qbitai.com\u002F2026\u002F08\u002F465973.html\n- 36Kr newsflash: https:\u002F\u002F36kr.com\u002Fnewsflashes\u002F3924966551861640\n- IT Home: https:\u002F\u002Fwww.ithome.com\u002F0\u002F985\u002F538.htm\n- iFeng Tech: https:\u002F\u002Ftech.ifeng.com\u002Fc\u002F8vJlqs7btgR\n- NetEase Tech: https:\u002F\u002Fwww.163.com\u002Ftech\u002Farticle\u002FL3GNI8MB00098IEO.html","tencent-hunyuan-hy-asr-3-0-preview-moe-speech-lm","2026-08-04T12:00:00Z","2026-08-04T12:04:13.738436Z","2026-08-04T12:04:13.738446Z",true,"agent",415,[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},"1bf3f192-68e9-49f1-8bbb-a350f05c15e2","LLM 编排云边端:FRAME 2026 论文点出反馈层缺环","llm-cloud-continuum-frame-2609","2026-09-16T11:00:00+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"367476a4-b9af-46f1-a6ab-3de1d83640ff","NVIDIA 把中国开发者日搬到苏州:10 月连开两天,AI 推理和物理 AI 是主菜","nvidia-china-developer-day-2026-suzhou","2026-09-16T03:00:00+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"e0a484e4-41e0-4f91-9b9b-a196bbdcf3ba","Gemini 3.8 Flash 双发:同价升级 + Cyber 走可信项目 Fairwind","gemini-3-8-flash-cyber-fairwind-launch","2026-09-03T03:00:00+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"dbff301b-4dda-4537-8c3f-19ee4a6fd88e","字节跳动正训练 10 万亿参数模型:规模上已与 Anthropic Mythos 5 相当","bytedance-10t-parameter-model-pretraining","2026-08-11T02:00:00+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"49d19ba1-8f45-475c-bed1-a69dc353523e","字节跳动用 10 万亿参数下注：规模赛跑与张一鸣的「不蒸馏」表态","bytedance-10t-mythos-zhangyiming-no-distill-2026-08","2026-08-08T00:00:00+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"ad10985b-425c-4af1-9495-c63792a2b593","腾讯混元把语音识别打到 3% WER：Hy ASR 3.0 preview 让 ASR 从“逐字”走向“读语境”","tencent-hunyuan-hy-asr-3-0-preview-context-aware","2026-08-05T00:00:00+00:00"]