随着chatgpt的火热，中文知识类问答数据集由收到诸多关注，其作为高质量的QA数据，可以用于SFT阶段以及pretrain预训练阶段。

本文主要介绍目前开源可下载的两个较大规模的知识类数据集，包括147万百度知道知识类数据集、425万社区问答webtext2019zh知识类数据2大数据集，以及8000条保险行业问答数据、15.6万条电信问答数据、77万条金融行业问答数据、20.3万条联通问答数据以及4万条农业银行问答数据等6大小规模领域问答数据集。

本项目充分借鉴了两个项目的工作，供大家一起参考。

一、147万百度知道知识类数据集

百度知道数据集含有1470142个预先过滤过的、高质量问题和答案，每个问题属于一个类别。总共有492个类别，其中频率达到或超过10次的类别有434个。

{"qid":,"category":,"title":,"desc":<desc>,"answer":<answer>}</code></pre><p><img>其中，category是问题的类型，title是问题的标题，desc是问题的描述，可以为空或与标题内容一致。</p><p><strong>数据样例</strong></p><pre><code>{
  "qid": "qid_6610724023825624555",
  "category": "娱乐-度假旅游",
  "title": "请问这起交通事故是谁的责任居多?小车和摩托车发生事故，在无红绿灯 ", 
  "desc": "小车和摩托车发生事故，在无红绿灯的十字路口，小停车看看左右，在觉得安全的情况下刹车慢慢以时速10公里左右的速度靠右行驶过路口，好没有出到十字路口正中时，被左边突然快速行驶过来的摩托车撞在车头前，  摩托车主摔到膝盖和檫伤脸部，请问这起交通事故是谁的责任居多。如果双方都有责任的话，大概各占几成？~\r",   "answer": "通过没有信号控制的十字路口，应该减速慢性，让右边的车先行，按你说的，摩托车好像在汽车的左边，所以严格来说可能摩托车全责。当然还要看汽车是否证照齐全，是否饮酒等。具体由交警调查后认定。"
},
{
  "qid": "qid_9099219392179923453",  
  "category": "教育/科学-理工学科-数学",
  "title": "一元一次方程!车间有28名工人,生产一种螺栓和螺帽,一个螺栓的两 ", "desc": "车间有28名工人,生产一种栓和螺帽,一个螺栓的两头各套上一个螺帽配成一套,每人每天可以生产螺栓12个或螺帽18个,问多少个工人生产螺栓,多少个工人生产螺帽,才能使一天所生产的螺栓和螺帽刚好配套??(要求用一元一次方程解)快!!",
  "answer": "设X人生产螺栓，那么生产螺帽的人就是28-X，\r\nX人可生产螺栓12X个，28-X人可生产螺帽18*（28-X）个，因为两个螺帽和一个螺栓配成一套，那么可列方程\r\n12X=18（28-X）/2\r\n12X=（504-18X）/2\r\n12X=252-9X\r\n12X+9X=252\r\n21X=252\r\nX=252/21=12人\r\n那么生产螺帽的就是28-X=28-12=16人。"
},
{
  "qid": "qid_3769366928615670292",
  "category": "生活-购物", "title": "开办网上购物网站，需要办理哪些手续？我想开办一个网上购物的网站， ", 
  "desc": "我想开办一个网上的网站，请问需要办理哪些手续？",
  "answer": "你的购物网站肯定要挂靠在公司下面\r\n先注册一个公司去吧"
}
{
  "qid": "qid_6490739535150628080",
  "category": "游戏-完美游戏-诛仙",
  "title": "08年28号新诛仙有电信新区吗?我想问下08年28号的新诛仙六道 ", 
  "desc": "我想问下08年28号的新六道轮回会不会开电信新区啊?御剑飞行系统是什么样的?", 
  "answer": "这个没有御剑飞行，好东西完美会慢慢出的。。。据说飞行速度比坐骑慢，不过是直线距离，还是划算，空中会有怪可以打。。坐骑嘛，完美这个钱还是要赚，所以他的速度应该会比飞行快，而且新出的会有属性加成。"
}</code></pre><p><img><strong>下载地址</strong>： https://pan.baidu.com/s/1XK3G8A3du2pme2-N5rtUVw 提取码: fgtf</p><h2>二、425万社区问答webtext2019zh知识类数据集</h2><p>社区问答webtext2019zh知识类数据集含有410万个预先过滤过的、高质量问题和回复。每个问题属于一个【话题】，总共有2.8万个各式话题，话题包罗万象。从1400万个原始问答中，筛选出至少获得3个点赞以上的的答案，代表了回复的内容比较不错或有趣，从而获得高质量的数据集。除了对每个问题对应一个话题、问题的描述、一个或多个回复外，每个回复还带有点赞数、回复ID、回复者的标签。</p><pre><code>{
    "qid":<qid>,
    "title":<title>,
    "desc":<desc>,
    "topic":<topic>,
    "star":<star>,
    "content":<content>,
    "answer_id":<answer_id>,
    "answerer_tags":<answerer_tags>
}</code></pre><p><img>其中:</p><p>qid是问题的id，title是问题的标题，desc是问题的描述，可以为空；topic是问题所属的话题，star是该回复的点赞个数，content是回复的内容;answer_id是回复的ID,answerer_tags是回复者所携带的标签。</p><p><strong>数据样例</strong></p><pre><code>{
  "qid": 20619381,
  "title": "张献忠血洗四川是否属实？",
  "desc": "", "topic": "明朝", 
  "star": 3, 
  "content": "四川人历史上有三次大灭绝，现在的川人基本都是湖广填四川填过来的，所以我认为这个基本属实。",
  "answer_id": 17447047,
  "answerer_tags": "如是我闻"
},
{
  "qid": 36651654, 
  "title": "你发现了哪些基于个人经验的神秘规律？",
  "desc": "One 里看到的，看看问答社区的朋友们有什么更加有趣的规律。",
  "topic": "经验",
  "star": 22,
  "content": "去吃饭的路比吃饭回来的路长",
  "answer_id": 112831136,
  "answerer_tags": "大盈若冲"
 },
 {
   "qid": 20801703, 
   "title": "为什么大多数楼盘的名字俗到不行？",
   "desc": "国内很多楼盘的名字都非常俗，像什么", 
   "topic": "房地产", 
   "star": 7,
   "content": "房子是卖给大众的，不是专卖给诗人文青的，我见过几个取的很雅的楼盘名字提案，都被毙掉了，因为你要解释这个名字就得一堆文字，不能给大众直观的感觉。另，现在觉得俗是因为见的多了。。。",
   "answer_id": 16245275, 
   "answerer_tags": "九丰投资"
  }</code></pre><p><img><strong>下载地址1：</strong>链接: https://pan.baidu.com/s/1V2iCtzX-XOuL1Mu1GbRCpA 提取码: n3r7</p><p><strong>下载地址2：</strong>https://drive.google.com/u/0/uc?id=1u2yW_XohbYL2YAK6Bzc5XrngHstQTf0v&export=download</p><h2>三、六大细分领域小规模知识问答数据集</h2><p>当然，出了大规模数据集外，还有较小规模的数据集，包括8000条保险行业问答数据、15.6万条电信问答数据、77万条金融行业问答数据、20.3万条联通问答数据以及4万条农业银行问答数据。</p><p><strong>1、8000条保险行业问答数据</strong></p><p><img alt="" src="https://img-blog.csdnimg.cn/img_convert/468e8770088b1b6abdfc4d0aaa3e906a.png"></p><p><strong>下载地址：</strong> https://pan.baidu.com/s/1cgYeIrJHAgb8D33H09Zc5w</p><p><strong>2、15.6万条电信问答数据</strong></p><p><img alt="" src="https://img-blog.csdnimg.cn/img_convert/71d2325f0a840b3421bdbc67861e6f38.png"></p><p><strong>下载地址：</strong> https://pan.baidu.com/s/1nrg5SRU3Xy1VN85dd85-vg</p><p><strong>3、77万条金融行业问答数据</strong></p><p><img alt="" src="https://img-blog.csdnimg.cn/img_convert/a643df940e1c3d188575f44f178140ad.png"><strong>下载地址：</strong> https://pan.baidu.com/s/1z1Rnnk-ubRSvzDu4UvLlIw</p><p><strong>4、3.6万条法律问答数据</strong></p><p><img alt="" src="https://img-blog.csdnimg.cn/img_convert/0c77a8964232bcce3c740e24feec17ec.png"></p><p><strong>下载地址：</strong> https://pan.baidu.com/s/18Lwq16VBo6wBD_qLb3i33g</p><p><strong>5、20.3万条联通问答数据 </strong></p><p><img alt="" src="https://img-blog.csdnimg.cn/img_convert/e0cbf5c23e9a60e436e59469f64491d8.png"></p><p><strong>下载地址：</strong> https://pan.baidu.com/s/1oYi9SfbXpnvreJYGV837Nw</p><p><strong>6、4万条农业银行问答数据</strong></p><p><img alt="" src="https://img-blog.csdnimg.cn/img_convert/202b7822433473ec6a9d0ec8e123e0ff.png"><strong>下载地址：</strong> https://pan.baidu.com/s/1n-jT9SKkt6cwI_PjCd7i_g</p><h2 style="text-align: center;"><font size="4">总结</font></h2><p>本文主要介绍了目前开源可下载的两个较大规模的知识类数据集，包括147万百度知道知识类数据集、425万社区问答webtext2019zh知识类数据2大数据集，以及8000条保险行业问答数据、15.6万条电信问答数据、77万条金融行业问答数据、20.3万条联通问答数据以及4万条农业银行问答数据等6大小规模领域问答数据集。</p><p>知识类QA数据集对于训练一个生成式的QA来说，具有一定的借鉴意义，大家可以基于该数据做更多尝试。</p><h2 style="text-align: center;"><font size="4">参考文献</font></h2><p>1、https://github.com/brightmart/nlp_chinese_corpus2、https://github.com/SophonPlus/ChineseNlpCorpus/</p></div><div class="mt-9"><h3 class="text-lg mb-5.5 text-1"> 评论 0</h3><div class="flex gap-3.5 mb-6 items-start"><img src="/img/avatars/default.png" class="w-9 h-9 rounded-full flex-shrink-0" alt="我的头像"><div class="flex-1"><textarea class="w-full text-sm leading-relaxed border border-base rounded-lg p-3.5 surface text-1 resize-y transition-all focus:outline-none focus:border-brand min-h-[120px]" placeholder="写下你的评论..."></textarea><div class="flex items-center justify-end mt-2.5"><span class="text-xs text-3 mr-auto">文明发言，友善讨论</span><button class="btn btn-primary btn-sm" disabled> 发表评论 </button></div></div></div><!--[--><!--]--><div class="text-center text-3 text-sm py-10 border-b border-base"> 还没有评论，发表你的看法，来抢沙发～ </div></div></div><aside class="flex flex-col gap-5"><div class="surface border border-base rounded-xl p-5.5 text-center"><img src="https://api.speechhome.com/Resources/Image/Avatar/2020-12-24/1282586712274505729.png" alt="语音小管家" class="w-20 h-20 rounded-full mx-auto mb-3 object-cover"><a href="/user/1282586712274505729" class="block text-base my-1.5 text-1 hover:text-brand">语音小管家</a><p class="text-xs text-3 m-0 mb-4">语音之家官方信息发布</p><button class="btn-primary btn btn-block">+ 关注</button></div><div class="surface border border-base rounded-xl p-4.5"><h3 class="text-base mb-3.5 flex items-center gap-2 text-1"><span class="w-1 h-4 bg-brand rounded"></span> 相关推荐 </h3><!--[--><a href="/blogs/opensource/2108490385755082752" class="flex gap-2.5 items-center surface border border-base rounded-lg p-2.5 mb-2 transition-all hover:border-[var(--color-brand-soft-2)]"><img src="https://api.speechhome.com/Resources/Thumb/2026-10-09/2108490380579311616.png" alt="Cactus Compute 开源 16.9MB 语音识别模型 Whistle" class="w-[94px] aspect-[2.35/1] object-cover rounded flex-shrink-0 surface-3"><div class="text-xs leading-snug text-1 text-[13px] min-w-0"><span class="line-clamp-2">Cactus Compute 开源 16.9MB 语音识别模型 Whistle</span><small class="block text-3 mt-0.5">2026-10-09</small></div></a><a href="/blogs/news/2108121611797401600" class="flex gap-2.5 items-center surface border border-base rounded-lg p-2.5 mb-2 transition-all hover:border-[var(--color-brand-soft-2)]"><img src="https://api.speechhome.com/Resources/Thumb/2026-10-08/2108121602066616320.png" alt="AudioEdit：深入探索基础模型时代的音频编辑" class="w-[94px] aspect-[2.35/1] object-cover rounded flex-shrink-0 surface-3"><div class="text-xs leading-snug text-1 text-[13px] min-w-0"><span class="line-clamp-2">AudioEdit：深入探索基础模型时代的音频编辑</span><small class="block text-3 mt-0.5">2026-10-08</small></div></a><a href="/blogs/news/2108125194785787904" class="flex gap-2.5 items-center surface border border-base rounded-lg p-2.5 mb-2 transition-all hover:border-[var(--color-brand-soft-2)]"><img src="https://api.speechhome.com/Resources/Thumb/2026-10-08/2108125140083675136.png" alt="[NCMMSC 2026] 青年论坛｜从语音大模型到音乐脑机接口，10场前沿报告齐聚横琴" class="w-[94px] aspect-[2.35/1] object-cover rounded flex-shrink-0 surface-3"><div class="text-xs leading-snug text-1 text-[13px] min-w-0"><span class="line-clamp-2">[NCMMSC 2026] 青年论坛｜从语音大模型到音乐脑机接口，10场前沿报告齐聚横琴</span><small class="block text-3 mt-0.5">2026-10-08</small></div></a><a href="/blogs/news/2108120734877487104" class="flex gap-2.5 items-center surface border border-base rounded-lg p-2.5 mb-2 transition-all hover:border-[var(--color-brand-soft-2)]"><img src="https://api.speechhome.com/Resources/Thumb/2026-10-08/2108121812306104320.jpg" alt="VoxMem：音频大模型能跨会话记住说话人、语气和环境声吗？" class="w-[94px] aspect-[2.35/1] object-cover rounded flex-shrink-0 surface-3"><div class="text-xs leading-snug text-1 text-[13px] min-w-0"><span class="line-clamp-2">VoxMem：音频大模型能跨会话记住说话人、语气和环境声吗？</span><small class="block text-3 mt-0.5">2026-10-06</small></div></a><a href="/blogs/algorithm/2108115606128693248" class="flex gap-2.5 items-center surface border border-base rounded-lg p-2.5 mb-2 transition-all hover:border-[var(--color-brand-soft-2)]"><img src="https://api.speechhome.com/Resources/Thumb/2026-10-08/2108115598969016320.png" alt="Edge0 发布中英双语流式语音识别模型 Audio8-ASR-Infinite" class="w-[94px] aspect-[2.35/1] object-cover rounded flex-shrink-0 surface-3"><div class="text-xs leading-snug text-1 text-[13px] min-w-0"><span class="line-clamp-2">Edge0 发布中英双语流式语音识别模型 Audio8-ASR-Infinite</span><small class="block text-3 mt-0.5">2026-09-30</small></div></a><a href="/blogs/news/2108115158432878592" class="flex gap-2.5 items-center surface border border-base rounded-lg p-2.5 mb-2 transition-all hover:border-[var(--color-brand-soft-2)]"><img src="https://api.speechhome.com/Resources/Thumb/2026-10-08/2108115105496567808.png" alt="活动报名：OpenAI 和 Thinking Machines 都押注的实时交互模型，下一步往哪走？丨iRTE 2026" class="w-[94px] aspect-[2.35/1] object-cover rounded flex-shrink-0 surface-3"><div class="text-xs leading-snug text-1 text-[13px] min-w-0"><span class="line-clamp-2">活动报名：OpenAI 和 Thinking Machines 都押注的实时交互模型，下一步往哪走？丨iRTE 2026</span><small class="block text-3 mt-0.5">2026-09-30</small></div></a><a href="/blogs/news/2103072719703773184" class="flex gap-2.5 items-center surface border border-base rounded-lg p-2.5 mb-2 transition-all hover:border-[var(--color-brand-soft-2)]"><img src="https://api.speechhome.com/Resources/Thumb/2026-09-24/2103072824741728256.png" alt="ICASSP 2027丨ADD 2026国际挑战赛开放报名" class="w-[94px] aspect-[2.35/1] object-cover rounded flex-shrink-0 surface-3"><div class="text-xs leading-snug text-1 text-[13px] min-w-0"><span class="line-clamp-2">ICASSP 2027丨ADD 2026国际挑战赛开放报名</span><small class="block text-3 mt-0.5">2026-09-24</small></div></a><a href="/blogs/news/2103070358063157248" class="flex gap-2.5 items-center surface border border-base rounded-lg p-2.5 mb-2 transition-all hover:border-[var(--color-brand-soft-2)]"><img src="https://api.speechhome.com/Resources/Thumb/2026-09-24/2103070353365536768.png" alt="ICASSP 2027丨赛事合集" class="w-[94px] aspect-[2.35/1] object-cover rounded flex-shrink-0 surface-3"><div class="text-xs leading-snug text-1 text-[13px] min-w-0"><span class="line-clamp-2">ICASSP 2027丨赛事合集</span><small class="block text-3 mt-0.5">2026-09-23</small></div></a><a href="/blogs/opensource/2103071658700050432" class="flex gap-2.5 items-center surface border border-base rounded-lg p-2.5 mb-2 transition-all hover:border-[var(--color-brand-soft-2)]"><img src="https://api.speechhome.com/Resources/Thumb/2026-09-24/2103071589976379392.png" alt="蚂蚁发布MECT声纹模型：MoE构架刷新VoxCeleb SOTA，并实现100ms流式声纹推理，代码模型已开源" class="w-[94px] aspect-[2.35/1] object-cover rounded flex-shrink-0 surface-3"><div class="text-xs leading-snug text-1 text-[13px] min-w-0"><span class="line-clamp-2">蚂蚁发布MECT声纹模型：MoE构架刷新VoxCeleb SOTA，并实现100ms流式声纹推理，代码模型已开源</span><small class="block text-3 mt-0.5">2026-09-23</small></div></a><a href="/blogs/algorithm/2103056363885301760" class="flex gap-2.5 items-center surface border border-base rounded-lg p-2.5 mb-2 transition-all hover:border-[var(--color-brand-soft-2)]"><img src="https://api.speechhome.com/Resources/Thumb/2026-09-24/2103056323817115648.png" alt="ISCSLP 2026丨ASLP@NPU&NJU实验室8篇录用论文分享" class="w-[94px] aspect-[2.35/1] object-cover rounded flex-shrink-0 surface-3"><div class="text-xs leading-snug text-1 text-[13px] min-w-0"><span class="line-clamp-2">ISCSLP 2026丨ASLP@NPU&NJU实验室8篇录用论文分享</span><small class="block text-3 mt-0.5">2026-09-23</small></div></a><!--]--></div><div class="surface border border-base rounded-xl p-4.5"><h3 class="text-base mb-3.5 flex items-center gap-2 text-1"><span class="w-1 h-4 bg-brand rounded"></span> 该作者的其他文章 </h3><!--[--><a href="/blogs/audiocc/2108872667896614912" class="flex items-center gap-2.5 py-2 border-b border-dashed border-base last:border-none text-sm text-2"><span class="w-5.5 h-5.5 rounded-md grid place-items-center flex-shrink-0 text-xs font-bold bg-brand-soft text-brand">1</span><span class="line-clamp-2 hover:text-brand">【交我学-43】多模态模型的“思考”方式：视觉、通用音频与语音的 CoT 技术路线</span></a><a href="/blogs/opensource/2108490385755082752" class="flex items-center gap-2.5 py-2 border-b border-dashed border-base last:border-none text-sm text-2"><span class="w-5.5 h-5.5 rounded-md grid place-items-center flex-shrink-0 text-xs font-bold bg-brand-soft text-brand">2</span><span class="line-clamp-2 hover:text-brand">Cactus Compute 开源 16.9MB 语音识别模型 Whistle</span></a><a href="/blogs/news/2108121611797401600" class="flex items-center gap-2.5 py-2 border-b border-dashed border-base last:border-none text-sm text-2"><span class="w-5.5 h-5.5 rounded-md grid place-items-center flex-shrink-0 text-xs font-bold bg-brand-soft text-brand">3</span><span class="line-clamp-2 hover:text-brand">AudioEdit：深入探索基础模型时代的音频编辑</span></a><a href="/blogs/news/2108125194785787904" class="flex items-center gap-2.5 py-2 border-b border-dashed border-base last:border-none text-sm text-2"><span class="w-5.5 h-5.5 rounded-md grid place-items-center flex-shrink-0 text-xs font-bold bg-brand-soft text-brand">4</span><span class="line-clamp-2 hover:text-brand">[NCMMSC 2026] 青年论坛｜从语音大模型到音乐脑机接口，10场前沿报告齐聚横琴</span></a><a href="/blogs/news/2108120734877487104" class="flex items-center gap-2.5 py-2 border-b border-dashed border-base last:border-none text-sm text-2"><span class="w-5.5 h-5.5 rounded-md grid place-items-center flex-shrink-0 text-xs font-bold bg-brand-soft text-brand">5</span><span class="line-clamp-2 hover:text-brand">VoxMem：音频大模型能跨会话记住说话人、语气和环境声吗？</span></a><a href="/blogs/algorithm/2108119561957150720" class="flex items-center gap-2.5 py-2 border-b border-dashed border-base last:border-none text-sm text-2"><span class="w-5.5 h-5.5 rounded-md grid place-items-center flex-shrink-0 text-xs font-bold bg-brand-soft text-brand">6</span><span class="line-clamp-2 hover:text-brand">EvoAudio：让音频大模型在“听力训练”中自进化</span></a><a href="/blogs/algorithm/2108115606128693248" class="flex items-center gap-2.5 py-2 border-b border-dashed border-base last:border-none text-sm text-2"><span class="w-5.5 h-5.5 rounded-md grid place-items-center flex-shrink-0 text-xs font-bold bg-brand-soft text-brand">7</span><span class="line-clamp-2 hover:text-brand">Edge0 发布中英双语流式语音识别模型 Audio8-ASR-Infinite</span></a><a href="/blogs/news/2108115158432878592" class="flex items-center gap-2.5 py-2 border-b border-dashed border-base last:border-none text-sm text-2"><span class="w-5.5 h-5.5 rounded-md grid place-items-center flex-shrink-0 text-xs font-bold bg-brand-soft text-brand">8</span><span class="line-clamp-2 hover:text-brand">活动报名：OpenAI 和 Thinking Machines 都押注的实时交互模型，下一步往哪走？丨iRTE 2026</span></a><a href="/blogs/opensource/2108114572035624960" class="flex items-center gap-2.5 py-2 border-b border-dashed border-base last:border-none text-sm text-2"><span class="w-5.5 h-5.5 rounded-md grid place-items-center flex-shrink-0 text-xs font-bold bg-brand-soft text-brand">9</span><span class="line-clamp-2 hover:text-brand">YODAS v3 开源 110 万小时 48kHz 多通道语音数据，覆盖 147 种语言</span></a><a href="/blogs/news/2103072719703773184" class="flex items-center gap-2.5 py-2 border-b border-dashed border-base last:border-none text-sm text-2"><span class="w-5.5 h-5.5 rounded-md grid place-items-center flex-shrink-0 text-xs font-bold bg-brand-soft text-brand">10</span><span class="line-clamp-2 hover:text-brand">ICASSP 2027丨ADD 2026国际挑战赛开放报名</span></a><!--]--></div></aside></div></div><!--]--></main><footer class="surface border-t border-base mt-10 py-6"><div class="container"><div class="grid grid-cols-2 md:grid-cols-[1.5fr_1.3fr_1.3fr_1fr] gap-6 items-start"><div class="col-span-2 md:col-span-1"><h4 class="text-[15px] mb-3 text-1 font-semibold"> 语音之家 SpeechHome </h4><p class="text-3 text-[13px] leading-relaxed m-0"> 汇聚产学研各界开发者，覆盖语音识别、音频合成、声纹识别、视听多模态、物理 AI、具身听觉、人机交互等方向。 </p></div><div><h4 class="text-[15px] mb-3 text-1 font-semibold">友情链接</h4><div class="grid grid-cols-2 gap-x-4.5"><!--[--><a href="https://www.ccf.org.cn/" target="_blank" rel="noopener noreferrer" class="block text-3 text-[13px] py-1 leading-relaxed hover:text-brand truncate">中国计算机学会</a><a href="http://www.aiiaorg.cn/" target="_blank" rel="noopener noreferrer" class="block text-3 text-[13px] py-1 leading-relaxed hover:text-brand truncate">中国人工智能产业发展联盟</a><a href="http://caai.cn/" target="_blank" rel="noopener noreferrer" class="block text-3 text-[13px] py-1 leading-relaxed hover:text-brand truncate">中国人工智能学会</a><a href="https://www.modelscope.cn/home?utm_source=kostja&utm_medium=blog" target="_blank" rel="noopener noreferrer" class="block text-3 text-[13px] py-1 leading-relaxed hover:text-brand truncate">魔搭社区</a><a href="http://www.aishelltech.com/" target="_blank" rel="noopener noreferrer" class="block text-3 text-[13px] py-1 leading-relaxed hover:text-brand truncate">希尔贝壳</a><a href=" https://www.oschina.net/" target="_blank" rel="noopener noreferrer" class="block text-3 text-[13px] py-1 leading-relaxed hover:text-brand truncate">开源中国</a><a href="https://segmentfault.com/" target="_blank" rel="noopener noreferrer" class="block text-3 text-[13px] py-1 leading-relaxed hover:text-brand truncate">思否</a><a href="https://juejin.cn/" target="_blank" rel="noopener noreferrer" class="block text-3 text-[13px] py-1 leading-relaxed hover:text-brand truncate">稀土掘金</a><a href="https://www.rtecommunity.dev/" target="_blank" rel="noopener noreferrer" class="block text-3 text-[13px] py-1 leading-relaxed hover:text-brand truncate">RTE开发者社区</a><a href="https://www.csdn.net/" target="_blank" rel="noopener noreferrer" class="block text-3 text-[13px] py-1 leading-relaxed hover:text-brand truncate">CSDN</a><a href="http://www.airadiance.cn/" target="_blank" rel="noopener noreferrer" class="block text-3 text-[13px] py-1 leading-relaxed hover:text-brand truncate">AI光影社 </a><a href="https://wenet.org.cn" target="_blank" rel="noopener noreferrer" class="block text-3 text-[13px] py-1 leading-relaxed hover:text-brand truncate">WeNet</a><a href="http://www.openslr.org/" target="_blank" rel="noopener noreferrer" class="block text-3 text-[13px] py-1 leading-relaxed hover:text-brand truncate">openslr</a><a href="https://mongoing.com/" target="_blank" rel="noopener noreferrer" class="block text-3 text-[13px] py-1 leading-relaxed hover:text-brand truncate">MongoDB中文社区</a><!--]--></div></div><div><h4 class="text-[15px] mb-3 text-1 font-semibold">资源</h4><div class="grid grid-cols-2 gap-x-4.5"><a href="/blogs" class="block text-3 text-[13px] py-1 leading-relaxed hover:text-brand"> 声浪 </a><a href="/events" class="block text-3 text-[13px] py-1 leading-relaxed hover:text-brand"> 活动 </a><a href="/aips" class="block text-3 text-[13px] py-1 leading-relaxed hover:text-brand"> AI科普 </a><a href="/share" class="block text-3 text-[13px] py-1 leading-relaxed hover:text-brand"> 分享 </a><a href="/jobs" class="block text-3 text-[13px] py-1 leading-relaxed hover:text-brand"> 热聘 </a><a href="/vip" class="block text-3 text-[13px] py-1 leading-relaxed hover:text-brand"> 会员服务 </a></div></div><div><h4 class="text-[15px] mb-3 text-1 font-semibold">关于</h4><a href="/other/terms" class="block text-3 text-[13px] py-1 leading-relaxed hover:text-brand"> 服务条款 </a><a href="/other/privacy" class="block text-3 text-[13px] py-1 leading-relaxed hover:text-brand"> 隐私协议 </a><a href="/partners" class="block text-3 text-[13px] py-1 leading-relaxed hover:text-brand"> 合作伙伴 </a></div></div><div class="border-t border-base mt-4 pt-3 text-center text-3 text-xs"><p class="m-0"><span>© 2026 语音之家 SpeechHome.com</span><span class="mx-2">|</span><a href="https://beian.miit.gov.cn/" target="_blank" rel="noopener noreferrer" class="hover:text-brand">京ICP备17027745号-4</a><span class="mx-2">|</span><a href="https://www.beian.gov.cn/portal/registerSystemInfo?recordcode=11010802032971" target="_blank" rel="noopener noreferrer" class="hover:text-brand">京公网安备11010802032971号</a></p></div></div></footer></div><span></span><span></span><!--]--></div><div id="teleports"></div><script>window.__NUXT__={};window.__NUXT__.config={public:{apiClientBase:"/api",wxLoginAppId:"",wxLoginRedirectUri:"",eduExternalUrl:"https://edu.speechhome.com",siteUrl:"https://speechhome.com",baiduAnalyticsId:"d295bb4c51448d20b740792d64b0dd49",piniaPluginPersistedstate:{cookieOptions:{maxAge:604800,path:"/"}}},app:{baseURL:"/",buildId:"433e7825-6524-4222-9543-424183ed611c",buildAssetsDir:"/_nuxt/",cdnURL:""}}</script><script type="application/json" data-nuxt-data="nuxt-app" data-ssr="true" id="__NUXT_DATA__">[["ShallowReactive",1],{"data":2,"state":239,"once":246,"_errors":247,"serverRendered":249,"path":250,"pinia":251},["ShallowReactive",3],{"friend-links":4,"post-1630094714327928832":79,"replies-1630094714327928832":108,"related-1630094714327928832":111,"author-posts-1630094714327928832":200},[5,15,20,25,30,35,40,44,49,54,59,64,69,74],{"id":6,"title":7,"url":8,"imgSrc":9,"carouselType":10,"startTime":11,"endTime":12,"sort":13,"target":13,"statusCode":14},"1571738828526456832","中国计算机学会","https:\u002F\u002Fwww.ccf.org.cn\u002F","",7,"2000-01-01 00:00:00","2050-01-01 00:00:00",1,0,{"id":16,"title":17,"url":18,"imgSrc":9,"carouselType":10,"startTime":11,"endTime":12,"sort":19,"target":13,"statusCode":14},"1571740806270816256","中国人工智能产业发展联盟","http:\u002F\u002Fwww.aiiaorg.cn\u002F",2,{"id":21,"title":22,"url":23,"imgSrc":9,"carouselType":10,"startTime":11,"endTime":12,"sort":24,"target":13,"statusCode":14},"1571741183930142720","中国人工智能学会","http:\u002F\u002Fcaai.cn\u002F",3,{"id":26,"title":27,"url":28,"imgSrc":9,"carouselType":10,"startTime":11,"endTime":12,"sort":29,"target":13,"statusCode":14},"1571741249650692096","魔搭社区","https:\u002F\u002Fwww.modelscope.cn\u002Fhome?utm_source=kostja&utm_medium=blog",4,{"id":31,"title":32,"url":33,"imgSrc":9,"carouselType":10,"startTime":11,"endTime":12,"sort":34,"target":13,"statusCode":14},"1571741548750704640","希尔贝壳","http:\u002F\u002Fwww.aishelltech.com\u002F",5,{"id":36,"title":37,"url":38,"imgSrc":9,"carouselType":10,"startTime":11,"endTime":12,"sort":39,"target":13,"statusCode":14},"1571741605864542208","开源中国"," https:\u002F\u002Fwww.oschina.net\u002F",6,{"id":41,"title":42,"url":43,"imgSrc":9,"carouselType":10,"startTime":11,"endTime":12,"sort":10,"target":13,"statusCode":14},"1571741777227026432","思否","https:\u002F\u002Fsegmentfault.com\u002F",{"id":45,"title":46,"url":47,"imgSrc":9,"carouselType":10,"startTime":11,"endTime":12,"sort":48,"target":13,"statusCode":14},"1571741857594085376","稀土掘金","https:\u002F\u002Fjuejin.cn\u002F",8,{"id":50,"title":51,"url":52,"imgSrc":9,"carouselType":10,"startTime":11,"endTime":12,"sort":53,"target":13,"statusCode":14},"1571741929337655296","RTE开发者社区","https:\u002F\u002Fwww.rtecommunity.dev\u002F",9,{"id":55,"title":56,"url":57,"imgSrc":9,"carouselType":10,"startTime":11,"endTime":12,"sort":58,"target":13,"statusCode":14},"1571741992008945664","CSDN","https:\u002F\u002Fwww.csdn.net\u002F",10,{"id":60,"title":61,"url":62,"imgSrc":9,"carouselType":10,"startTime":11,"endTime":12,"sort":63,"target":13,"statusCode":14},"1571742099206967296","AI光影社 ","http:\u002F\u002Fwww.airadiance.cn\u002F",11,{"id":65,"title":66,"url":67,"imgSrc":9,"carouselType":10,"startTime":11,"endTime":12,"sort":68,"target":13,"statusCode":14},"1571742176633819136","WeNet","https:\u002F\u002Fwenet.org.cn",12,{"id":70,"title":71,"url":72,"imgSrc":9,"carouselType":10,"startTime":11,"endTime":12,"sort":73,"target":13,"statusCode":14},"1571742254131974144","openslr","http:\u002F\u002Fwww.openslr.org\u002F",13,{"id":75,"title":76,"url":77,"imgSrc":9,"carouselType":10,"startTime":11,"endTime":12,"sort":78,"target":13,"statusCode":14},"1571743803889553408","MongoDB中文社区","https:\u002F\u002Fmongoing.com\u002F",15,{"post":80,"message":9},{"id":81,"title":82,"author":83,"content":101,"postType":24,"subjectId":102,"subjectIds":93,"tag":9,"coverImage":103,"coverImageRelativePath":93,"readingCount":104,"creator":84,"olink":9,"workPlace":93,"salaryRange":93,"createTime":105,"likeCount":14,"collectCount":14,"replyCount":14,"isLike":106,"isCollect":106,"publishDate":107,"subjectEngName":93,"statusCode":14},"1630094714327928832","知识类问答数据集资源对外开放：百万级百度知道、社区问答及六大领域级小规模语料概述",{"id":84,"account":85,"name":86,"level":87,"position":88,"avatar":89,"company":90,"role":24,"introduction":91,"experience":14,"readCount":14,"likeCount":14,"followCount":14,"fansCount":14,"postCount":14,"statusCode":14,"phone":92,"email":93,"follow":93,"createTime":94,"studentID":93,"weChat":9,"hwClassName":93,"traces":93,"jobStatus":95,"country":96,"city":97,"graduatingInstitution":93,"graduationYear":14,"currentInstitution":93,"major":93,"degree":93,"experienceYears":93,"realName":86,"suggestion":93,"employer":93,"employerType":93,"membershipNumber":98,"other":93,"captcha":93,"modifyTime":99,"page":14,"size":14,"province":100,"regFrom":93,"vipName":93,"timeExpire":93},"1282586712274505729","speechhome","语音小管家",999,"官方小助手","https:\u002F\u002Fapi.speechhome.com\u002FResources\u002FImage\u002FAvatar\u002F2020-12-24\u002F1282586712274505729.png","语音之家","语音之家官方信息发布","0",null,"2020-07-13T16:04:36.17202","其他","中国","北京市","SHP000001","2024-08-01T16:59:58.726334","北京"," \u003Cp>随着chatgpt的火热，中文知识类问答数据集由收到诸多关注，其作为高质量的QA数据，可以用于SFT阶段以及pretrain预训练阶段。\u003Cbr>\u003C\u002Fp>\u003Cp>本文主要介绍目前开源可下载的两个较大规模的知识类数据集，包括147万百度知道知识类数据集、425万社区问答webtext2019zh知识类数据2大数据集，以及8000条保险行业问答数据、15.6万条电信问答数据、77万条金融行业问答数据、20.3万条联通问答数据以及4万条农业银行问答数据等6大小规模领域问答数据集。\u003C\u002Fp>\u003Cp>本项目充分借鉴了两个项目的工作，供大家一起参考。\u003C\u002Fp>\u003Ch2>一、147万百度知道知识类数据集\u003C\u002Fh2>\u003Cp>百度知道数据集含有1470142个预先过滤过的、高质量问题和答案，每个问题属于一个类别。总共有492个类别，其中频率达到或超过10次的类别有434个。\u003C\u002Fp>\u003Cpre>\u003Ccode>{\"qid\":\u003Cqid>,\"category\":\u003Ccategory>,\"title\":\u003Ctitle>,\"desc\":\u003Cdesc>,\"answer\":\u003Canswer>}\u003C\u002Fcode>\u003C\u002Fpre>\u003Cp>\u003Cimg src=\"data:image\u002Fgif;base64,R0lGODlhAQABAPABAP\u002F\u002F\u002FwAAACH5BAEKAAAALAAAAAABAAEAAAICRAEAOw==\">其中，category是问题的类型，title是问题的标题，desc是问题的描述，可以为空或与标题内容一致。\u003C\u002Fp>\u003Cp>\u003Cstrong>数据样例\u003C\u002Fstrong>\u003C\u002Fp>\u003Cpre>\u003Ccode>{\n  \"qid\": \"qid_6610724023825624555\",\n  \"category\": \"娱乐-度假旅游\",\n  \"title\": \"请问这起交通事故是谁的责任居多?小车和摩托车发生事故，在无红绿灯 \", \n  \"desc\": \"小车和摩托车发生事故，在无红绿灯的十字路口，小停车看看左右，在觉得安全的情况下刹车慢慢以时速10公里左右的速度靠右行驶过路口，好没有出到十字路口正中时，被左边突然快速行驶过来的摩托车撞在车头前，  摩托车主摔到膝盖和檫伤脸部，请问这起交通事故是谁的责任居多。如果双方都有责任的话，大概各占几成？~\\r\",   \"answer\": \"通过没有信号控制的十字路口，应该减速慢性，让右边的车先行，按你说的，摩托车好像在汽车的左边，所以严格来说可能摩托车全责。当然还要看汽车是否证照齐全，是否饮酒等。具体由交警调查后认定。\"\n},\n{\n  \"qid\": \"qid_9099219392179923453\",  \n  \"category\": \"教育\u002F科学-理工学科-数学\",\n  \"title\": \"一元一次方程!车间有28名工人,生产一种螺栓和螺帽,一个螺栓的两 \", \"desc\": \"车间有28名工人,生产一种栓和螺帽,一个螺栓的两头各套上一个螺帽配成一套,每人每天可以生产螺栓12个或螺帽18个,问多少个工人生产螺栓,多少个工人生产螺帽,才能使一天所生产的螺栓和螺帽刚好配套??(要求用一元一次方程解)快!!\",\n  \"answer\": \"设X人生产螺栓，那么生产螺帽的人就是28-X，\\r\\nX人可生产螺栓12X个，28-X人可生产螺帽18*（28-X）个，因为两个螺帽和一个螺栓配成一套，那么可列方程\\r\\n12X=18（28-X）\u002F2\\r\\n12X=（504-18X）\u002F2\\r\\n12X=252-9X\\r\\n12X+9X=252\\r\\n21X=252\\r\\nX=252\u002F21=12人\\r\\n那么生产螺帽的就是28-X=28-12=16人。\"\n},\n{\n  \"qid\": \"qid_3769366928615670292\",\n  \"category\": \"生活-购物\", \"title\": \"开办网上购物网站，需要办理哪些手续？我想开办一个网上购物的网站， \", \n  \"desc\": \"我想开办一个网上的网站，请问需要办理哪些手续？\",\n  \"answer\": \"你的购物网站肯定要挂靠在公司下面\\r\\n先注册一个公司去吧\"\n}\n{\n  \"qid\": \"qid_6490739535150628080\",\n  \"category\": \"游戏-完美游戏-诛仙\",\n  \"title\": \"08年28号新诛仙有电信新区吗?我想问下08年28号的新诛仙六道 \", \n  \"desc\": \"我想问下08年28号的新六道轮回会不会开电信新区啊?御剑飞行系统是什么样的?\", \n  \"answer\": \"这个没有御剑飞行，好东西完美会慢慢出的。。。据说飞行速度比坐骑慢，不过是直线距离，还是划算，空中会有怪可以打。。坐骑嘛，完美这个钱还是要赚，所以他的速度应该会比飞行快，而且新出的会有属性加成。\"\n}\u003C\u002Fcode>\u003C\u002Fpre>\u003Cp>\u003Cimg src=\"data:image\u002Fgif;base64,R0lGODlhAQABAPABAP\u002F\u002F\u002FwAAACH5BAEKAAAALAAAAAABAAEAAAICRAEAOw==\">\u003Cstrong>下载地址\u003C\u002Fstrong>： https:\u002F\u002Fpan.baidu.com\u002Fs\u002F1XK3G8A3du2pme2-N5rtUVw 提取码: fgtf\u003C\u002Fp>\u003Ch2>二、425万社区问答webtext2019zh知识类数据集\u003C\u002Fh2>\u003Cp>社区问答webtext2019zh知识类数据集含有410万个预先过滤过的、高质量问题和回复。每个问题属于一个【话题】，总共有2.8万个各式话题，话题包罗万象。从1400万个原始问答中，筛选出至少获得3个点赞以上的的答案，代表了回复的内容比较不错或有趣，从而获得高质量的数据集。除了对每个问题对应一个话题、问题的描述、一个或多个回复外，每个回复还带有点赞数、回复ID、回复者的标签。\u003C\u002Fp>\u003Cpre>\u003Ccode>{\n    \"qid\":\u003Cqid>,\n    \"title\":\u003Ctitle>,\n    \"desc\":\u003Cdesc>,\n    \"topic\":\u003Ctopic>,\n    \"star\":\u003Cstar>,\n    \"content\":\u003Ccontent>,\n    \"answer_id\":\u003Canswer_id>,\n    \"answerer_tags\":\u003Canswerer_tags>\n}\u003C\u002Fcode>\u003C\u002Fpre>\u003Cp>\u003Cimg src=\"data:image\u002Fgif;base64,R0lGODlhAQABAPABAP\u002F\u002F\u002FwAAACH5BAEKAAAALAAAAAABAAEAAAICRAEAOw==\">其中:\u003C\u002Fp>\u003Cp>qid是问题的id，title是问题的标题，desc是问题的描述，可以为空；topic是问题所属的话题，star是该回复的点赞个数，content是回复的内容;answer_id是回复的ID,answerer_tags是回复者所携带的标签。\u003C\u002Fp>\u003Cp>\u003Cstrong>数据样例\u003C\u002Fstrong>\u003C\u002Fp>\u003Cpre>\u003Ccode>{\n  \"qid\": 20619381,\n  \"title\": \"张献忠血洗四川是否属实？\",\n  \"desc\": \"\", \"topic\": \"明朝\", \n  \"star\": 3, \n  \"content\": \"四川人历史上有三次大灭绝，现在的川人基本都是湖广填四川填过来的，所以我认为这个基本属实。\",\n  \"answer_id\": 17447047,\n  \"answerer_tags\": \"如是我闻\"\n},\n{\n  \"qid\": 36651654, \n  \"title\": \"你发现了哪些基于个人经验的神秘规律？\",\n  \"desc\": \"One 里看到的，看看问答社区的朋友们有什么更加有趣的规律。\",\n  \"topic\": \"经验\",\n  \"star\": 22,\n  \"content\": \"去吃饭的路比吃饭回来的路长\",\n  \"answer_id\": 112831136,\n  \"answerer_tags\": \"大盈若冲\"\n },\n {\n   \"qid\": 20801703, \n   \"title\": \"为什么大多数楼盘的名字俗到不行？\",\n   \"desc\": \"国内很多楼盘的名字都非常俗，像什么\", \n   \"topic\": \"房地产\", \n   \"star\": 7,\n   \"content\": \"房子是卖给大众的，不是专卖给诗人文青的，我见过几个取的很雅的楼盘名字提案，都被毙掉了，因为你要解释这个名字就得一堆文字，不能给大众直观的感觉。另，现在觉得俗是因为见的多了。。。\",\n   \"answer_id\": 16245275, \n   \"answerer_tags\": \"九丰投资\"\n  }\u003C\u002Fcode>\u003C\u002Fpre>\u003Cp>\u003Cimg src=\"data:image\u002Fgif;base64,R0lGODlhAQABAPABAP\u002F\u002F\u002FwAAACH5BAEKAAAALAAAAAABAAEAAAICRAEAOw==\">\u003Cstrong>下载地址1：\u003C\u002Fstrong>链接: https:\u002F\u002Fpan.baidu.com\u002Fs\u002F1V2iCtzX-XOuL1Mu1GbRCpA 提取码: n3r7\u003C\u002Fp>\u003Cp>\u003Cstrong>下载地址2：\u003C\u002Fstrong>https:\u002F\u002Fdrive.google.com\u002Fu\u002F0\u002Fuc?id=1u2yW_XohbYL2YAK6Bzc5XrngHstQTf0v&export=download\u003C\u002Fp>\u003Ch2>三、六大细分领域小规模知识问答数据集\u003C\u002Fh2>\u003Cp>当然，出了大规模数据集外，还有较小规模的数据集，包括8000条保险行业问答数据、15.6万条电信问答数据、77万条金融行业问答数据、20.3万条联通问答数据以及4万条农业银行问答数据。\u003C\u002Fp>\u003Cp>\u003Cstrong>1、8000条保险行业问答数据\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp>\u003Cimg alt=\"\" src=\"https:\u002F\u002Fimg-blog.csdnimg.cn\u002Fimg_convert\u002F468e8770088b1b6abdfc4d0aaa3e906a.png\">\u003C\u002Fp>\u003Cp>\u003Cstrong>下载地址：\u003C\u002Fstrong> https:\u002F\u002Fpan.baidu.com\u002Fs\u002F1cgYeIrJHAgb8D33H09Zc5w\u003C\u002Fp>\u003Cp>\u003Cstrong>2、15.6万条电信问答数据\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp>\u003Cimg alt=\"\" src=\"https:\u002F\u002Fimg-blog.csdnimg.cn\u002Fimg_convert\u002F71d2325f0a840b3421bdbc67861e6f38.png\">\u003C\u002Fp>\u003Cp>\u003Cstrong>下载地址：\u003C\u002Fstrong> https:\u002F\u002Fpan.baidu.com\u002Fs\u002F1nrg5SRU3Xy1VN85dd85-vg\u003C\u002Fp>\u003Cp>\u003Cstrong>3、77万条金融行业问答数据\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp>\u003Cimg alt=\"\" src=\"https:\u002F\u002Fimg-blog.csdnimg.cn\u002Fimg_convert\u002Fa643df940e1c3d188575f44f178140ad.png\">\u003Cstrong>下载地址：\u003C\u002Fstrong> https:\u002F\u002Fpan.baidu.com\u002Fs\u002F1z1Rnnk-ubRSvzDu4UvLlIw\u003C\u002Fp>\u003Cp>\u003Cstrong>4、3.6万条法律问答数据\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp>\u003Cimg alt=\"\" src=\"https:\u002F\u002Fimg-blog.csdnimg.cn\u002Fimg_convert\u002F0c77a8964232bcce3c740e24feec17ec.png\">\u003C\u002Fp>\u003Cp>\u003Cstrong>下载地址：\u003C\u002Fstrong> https:\u002F\u002Fpan.baidu.com\u002Fs\u002F18Lwq16VBo6wBD_qLb3i33g\u003C\u002Fp>\u003Cp>\u003Cstrong>5、20.3万条联通问答数据 \u003C\u002Fstrong>\u003C\u002Fp>\u003Cp>\u003Cimg alt=\"\" src=\"https:\u002F\u002Fimg-blog.csdnimg.cn\u002Fimg_convert\u002Fe0cbf5c23e9a60e436e59469f64491d8.png\">\u003C\u002Fp>\u003Cp>\u003Cstrong>下载地址：\u003C\u002Fstrong> https:\u002F\u002Fpan.baidu.com\u002Fs\u002F1oYi9SfbXpnvreJYGV837Nw\u003C\u002Fp>\u003Cp>\u003Cstrong>6、4万条农业银行问答数据\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp>\u003Cimg alt=\"\" src=\"https:\u002F\u002Fimg-blog.csdnimg.cn\u002Fimg_convert\u002F202b7822433473ec6a9d0ec8e123e0ff.png\">\u003Cstrong>下载地址：\u003C\u002Fstrong> https:\u002F\u002Fpan.baidu.com\u002Fs\u002F1n-jT9SKkt6cwI_PjCd7i_g\u003C\u002Fp>\u003Ch2 style=\"text-align: center;\">\u003Cfont size=\"4\">总结\u003C\u002Ffont>\u003C\u002Fh2>\u003Cp>本文主要介绍了目前开源可下载的两个较大规模的知识类数据集，包括147万百度知道知识类数据集、425万社区问答webtext2019zh知识类数据2大数据集，以及8000条保险行业问答数据、15.6万条电信问答数据、77万条金融行业问答数据、20.3万条联通问答数据以及4万条农业银行问答数据等6大小规模领域问答数据集。\u003C\u002Fp>\u003Cp>知识类QA数据集对于训练一个生成式的QA来说，具有一定的借鉴意义，大家可以基于该数据做更多尝试。\u003C\u002Fp>\u003Ch2 style=\"text-align: center;\">\u003Cfont size=\"4\">参考文献\u003C\u002Ffont>\u003C\u002Fh2>\u003Cp>1、https:\u002F\u002Fgithub.com\u002Fbrightmart\u002Fnlp_chinese_corpus2、https:\u002F\u002Fgithub.com\u002FSophonPlus\u002FChineseNlpCorpus\u002F\u003C\u002Fp>","2","https:\u002F\u002Fapi.speechhome.com\u002FResources\u002FImage\u002F2023-02-27\u002F1630094705901572096_mark.png",100,"2023-02-27T14:27:03",false,"0001-01-01T00:00:00",{"indexFrom":13,"pageIndex":13,"pageSize":109,"totalCount":14,"totalPages":14,"items":110,"hasPreviousPage":106,"hasNextPage":106},20,[],[112,123,133,141,149,160,168,176,184,192],{"id":113,"title":114,"subjectId":115,"subjectIds":116,"type":19,"author":117,"tag":9,"createTime":118,"likeCount":14,"replyCount":14,"readingCount":119,"coverImage":120,"coverThumbPath":93,"describe":121,"statusCode":14,"sort":14,"subjectEngName":122,"workPlace":9,"salaryRange":9},"2108490385755082752","Cactus Compute 开源 16.9MB 语音识别模型 Whistle","4","4,2",{"id":84,"account":93,"name":86,"level":87,"position":88,"avatar":89,"company":93,"role":93,"introduction":93,"experience":14,"readCount":14,"likeCount":14,"followCount":14,"fansCount":14,"postCount":14,"statusCode":93,"phone":93,"email":93,"follow":93,"createTime":107,"studentID":93,"weChat":93,"hwClassName":93,"traces":93,"jobStatus":93,"country":93,"city":93,"graduatingInstitution":93,"graduationYear":93,"currentInstitution":93,"major":93,"degree":93,"experienceYears":93,"realName":93,"suggestion":93,"employer":93,"employerType":93,"membershipNumber":93,"other":93,"captcha":93,"modifyTime":93,"page":14,"size":14,"province":93,"regFrom":93,"vipName":93,"timeExpire":93},"2026-10-09T17:30:00",75,"https:\u002F\u002Fapi.speechhome.com\u002FResources\u002FThumb\u002F2026-10-09\u002F2108490380579311616.png","Cactus Compute 发布开源语音识别模型 Whistle，面向手机、可穿戴设备、机器人、汽车和微控制器等端侧设备。 模型文件仅 16.9MB，直接在 CPU 上运行，并与其端侧工具调用模型 Needle 共用同一套 C++ 推理引擎，可以把语音转写和后续工具调用放进同一个本地执行链路。 模…","opensource",{"id":124,"title":125,"subjectId":102,"subjectIds":126,"type":19,"author":127,"tag":9,"createTime":128,"likeCount":14,"replyCount":14,"readingCount":129,"coverImage":130,"coverThumbPath":93,"describe":131,"statusCode":14,"sort":14,"subjectEngName":132,"workPlace":9,"salaryRange":9},"2108121611797401600","AudioEdit：深入探索基础模型时代的音频编辑","2,5",{"id":84,"account":93,"name":86,"level":87,"position":88,"avatar":89,"company":93,"role":93,"introduction":93,"experience":14,"readCount":14,"likeCount":14,"followCount":14,"fansCount":14,"postCount":14,"statusCode":93,"phone":93,"email":93,"follow":93,"createTime":107,"studentID":93,"weChat":93,"hwClassName":93,"traces":93,"jobStatus":93,"country":93,"city":93,"graduatingInstitution":93,"graduationYear":93,"currentInstitution":93,"major":93,"degree":93,"experienceYears":93,"realName":93,"suggestion":93,"employer":93,"employerType":93,"membershipNumber":93,"other":93,"captcha":93,"modifyTime":93,"page":14,"size":14,"province":93,"regFrom":93,"vipName":93,"timeExpire":93},"2026-10-08T17:02:00",59,"https:\u002F\u002Fapi.speechhome.com\u002FResources\u002FThumb\u002F2026-10-08\u002F2108121602066616320.png","把录音里说错的一个词改过来，让一句话的语气更轻松，或者去掉背景中的狗叫声，同时保留原本的人声和环境声——这些看似简单的要求，正是音频编辑需要解决的问题。 随着音频生成模型的发展，用文字描述一段声音并让模型生成出来，已经有了不少可行方案。但面对一段已有的录音，模型还需要知道改什么、在哪里改，以及哪些内…","news",{"id":134,"title":135,"subjectId":102,"subjectIds":102,"type":29,"author":136,"tag":9,"createTime":137,"likeCount":14,"replyCount":14,"readingCount":138,"coverImage":139,"coverThumbPath":93,"describe":140,"statusCode":14,"sort":14,"subjectEngName":132,"workPlace":9,"salaryRange":9},"2108125194785787904","[NCMMSC 2026] 青年论坛｜从语音大模型到音乐脑机接口，10场前沿报告齐聚横琴",{"id":84,"account":93,"name":86,"level":87,"position":88,"avatar":89,"company":93,"role":93,"introduction":93,"experience":14,"readCount":14,"likeCount":14,"followCount":14,"fansCount":14,"postCount":14,"statusCode":93,"phone":93,"email":93,"follow":93,"createTime":107,"studentID":93,"weChat":93,"hwClassName":93,"traces":93,"jobStatus":93,"country":93,"city":93,"graduatingInstitution":93,"graduationYear":93,"currentInstitution":93,"major":93,"degree":93,"experienceYears":93,"realName":93,"suggestion":93,"employer":93,"employerType":93,"membershipNumber":93,"other":93,"captcha":93,"modifyTime":93,"page":14,"size":14,"province":93,"regFrom":93,"vipName":93,"timeExpire":93},"2026-10-08T08:07:00",87,"https:\u002F\u002Fapi.speechhome.com\u002FResources\u002FThumb\u002F2026-10-08\u002F2108125140083675136.png","声学智能的前沿探索与青年力量 当语音技术不再只是“听清楚、说出来”，当声音开始与大模型、多模态智能、医疗健康、脑科学、低资源语言深度交叉，我们正在进入一个更加丰富的声学智能时代。 从低码率语音编解码、语音增强与合成，到音频推理、副语言表达与障碍语音识别；从音乐脑机接口、认知健康智能诊断，到多模态小语…",{"id":142,"title":143,"subjectId":102,"subjectIds":126,"type":19,"author":144,"tag":9,"createTime":145,"likeCount":14,"replyCount":14,"readingCount":146,"coverImage":147,"coverThumbPath":93,"describe":148,"statusCode":14,"sort":14,"subjectEngName":132,"workPlace":9,"salaryRange":9},"2108120734877487104","VoxMem：音频大模型能跨会话记住说话人、语气和环境声吗？",{"id":84,"account":93,"name":86,"level":87,"position":88,"avatar":89,"company":93,"role":93,"introduction":93,"experience":14,"readCount":14,"likeCount":14,"followCount":14,"fansCount":14,"postCount":14,"statusCode":93,"phone":93,"email":93,"follow":93,"createTime":107,"studentID":93,"weChat":93,"hwClassName":93,"traces":93,"jobStatus":93,"country":93,"city":93,"graduatingInstitution":93,"graduationYear":93,"currentInstitution":93,"major":93,"degree":93,"experienceYears":93,"realName":93,"suggestion":93,"employer":93,"employerType":93,"membershipNumber":93,"other":93,"captcha":93,"modifyTime":93,"page":14,"size":14,"province":93,"regFrom":93,"vipName":93,"timeExpire":93},"2026-10-06T18:57:00",61,"https:\u002F\u002Fapi.speechhome.com\u002FResources\u002FThumb\u002F2026-10-08\u002F2108121812306104320.jpg","语音助手正在从“一问一答”走向跨越多次会话的长期交互。用户上周提过的计划、某位家庭成员说过的话、某次对话时的语气，都可能在之后被问起。但语音里的信息远不止文字：谁说的、怎么说的、背景里有什么声音，都只存在于音频信号里，一旦转成文字就丢失了。 来自墨尔本大学和新南威尔士大学的研究者提出了VoxMem，…",{"id":150,"title":151,"subjectId":152,"subjectIds":153,"type":19,"author":154,"tag":9,"createTime":155,"likeCount":14,"replyCount":14,"readingCount":156,"coverImage":157,"coverThumbPath":93,"describe":158,"statusCode":14,"sort":14,"subjectEngName":159,"workPlace":9,"salaryRange":9},"2108115606128693248","Edge0 发布中英双语流式语音识别模型 Audio8-ASR-Infinite","5","5,2",{"id":84,"account":93,"name":86,"level":87,"position":88,"avatar":89,"company":93,"role":93,"introduction":93,"experience":14,"readCount":14,"likeCount":14,"followCount":14,"fansCount":14,"postCount":14,"statusCode":93,"phone":93,"email":93,"follow":93,"createTime":107,"studentID":93,"weChat":93,"hwClassName":93,"traces":93,"jobStatus":93,"country":93,"city":93,"graduatingInstitution":93,"graduationYear":93,"currentInstitution":93,"major":93,"degree":93,"experienceYears":93,"realName":93,"suggestion":93,"employer":93,"employerType":93,"membershipNumber":93,"other":93,"captcha":93,"modifyTime":93,"page":14,"size":14,"province":93,"regFrom":93,"vipName":93,"timeExpire":93},"2026-09-30T19:40:00",47,"https:\u002F\u002Fapi.speechhome.com\u002FResources\u002FThumb\u002F2026-10-08\u002F2108115598969016320.png","Edge0 发布 Audio8-ASR-Infinite 预览版，面向中英文实时语音识别，提供 80、120 和 160ms 三档流式音频时钟，并可将转写延迟配置在 240–560ms。 模型原生上下文为 30 秒，但通过滚动 KV 缓存持续淘汰旧状态，可在内存占用和推理延迟不随音频时长增长的情况下…","algorithm",{"id":161,"title":162,"subjectId":102,"subjectIds":126,"type":29,"author":163,"tag":9,"createTime":164,"likeCount":14,"replyCount":14,"readingCount":165,"coverImage":166,"coverThumbPath":93,"describe":167,"statusCode":14,"sort":14,"subjectEngName":132,"workPlace":9,"salaryRange":9},"2108115158432878592","活动报名：OpenAI 和 Thinking Machines 都押注的实时交互模型，下一步往哪走？丨iRTE 2026",{"id":84,"account":93,"name":86,"level":87,"position":88,"avatar":89,"company":93,"role":93,"introduction":93,"experience":14,"readCount":14,"likeCount":14,"followCount":14,"fansCount":14,"postCount":14,"statusCode":93,"phone":93,"email":93,"follow":93,"createTime":107,"studentID":93,"weChat":93,"hwClassName":93,"traces":93,"jobStatus":93,"country":93,"city":93,"graduatingInstitution":93,"graduationYear":93,"currentInstitution":93,"major":93,"degree":93,"experienceYears":93,"realName":93,"suggestion":93,"employer":93,"employerType":93,"membershipNumber":93,"other":93,"captcha":93,"modifyTime":93,"page":14,"size":14,"province":93,"regFrom":93,"vipName":93,"timeExpire":93},"2026-09-30T16:37:00",52,"https:\u002F\u002Fapi.speechhome.com\u002FResources\u002FThumb\u002F2026-10-08\u002F2108115105496567808.png","过去一年，OpenAI、Thinking Machines Lab 等多模态团队开始把注意力从「生成更拟人的语音回答」，迈向一个更难的问题：AI 如何真正参与一场实时发生的互动。 全双工模型让智能体边听边说，全模态让语音、视觉等信息协同判断，而交互模型又往前一步：此刻这场互动正在发生什么，下一步应该…",{"id":169,"title":170,"subjectId":102,"subjectIds":102,"type":29,"author":171,"tag":9,"createTime":172,"likeCount":14,"replyCount":14,"readingCount":173,"coverImage":174,"coverThumbPath":93,"describe":175,"statusCode":14,"sort":14,"subjectEngName":132,"workPlace":9,"salaryRange":9},"2103072719703773184","ICASSP 2027丨ADD 2026国际挑战赛开放报名",{"id":84,"account":93,"name":86,"level":87,"position":88,"avatar":89,"company":93,"role":93,"introduction":93,"experience":14,"readCount":14,"likeCount":14,"followCount":14,"fansCount":14,"postCount":14,"statusCode":93,"phone":93,"email":93,"follow":93,"createTime":107,"studentID":93,"weChat":93,"hwClassName":93,"traces":93,"jobStatus":93,"country":93,"city":93,"graduatingInstitution":93,"graduationYear":93,"currentInstitution":93,"major":93,"degree":93,"experienceYears":93,"realName":93,"suggestion":93,"employer":93,"employerType":93,"membershipNumber":93,"other":93,"captcha":93,"modifyTime":93,"page":14,"size":14,"province":93,"regFrom":93,"vipName":93,"timeExpire":93},"2026-09-24T18:41:00",60,"https:\u002F\u002Fapi.speechhome.com\u002FResources\u002FThumb\u002F2026-09-24\u002F2103072824741728256.png","虚假音频检测技术日益受到关注，但现有技术仍存在两方面的局限：一是人工模仿音频检测。真人模仿者可复现目标说话人的韵律、音色和风格，与AI合成音频不同，人工模仿音频无明显生成痕迹，对现有检测方法构成严峻挑战。二是可解释性检测。为增强检测可信度，需明确判别依据与理由，而当前相关研究尚不够深入。为此，继 A…",{"id":177,"title":178,"subjectId":102,"subjectIds":102,"type":19,"author":179,"tag":9,"createTime":180,"likeCount":14,"replyCount":14,"readingCount":181,"coverImage":182,"coverThumbPath":93,"describe":183,"statusCode":14,"sort":14,"subjectEngName":132,"workPlace":9,"salaryRange":9},"2103070358063157248","ICASSP 2027丨赛事合集",{"id":84,"account":93,"name":86,"level":87,"position":88,"avatar":89,"company":93,"role":93,"introduction":93,"experience":14,"readCount":14,"likeCount":14,"followCount":14,"fansCount":14,"postCount":14,"statusCode":93,"phone":93,"email":93,"follow":93,"createTime":107,"studentID":93,"weChat":93,"hwClassName":93,"traces":93,"jobStatus":93,"country":93,"city":93,"graduatingInstitution":93,"graduationYear":93,"currentInstitution":93,"major":93,"degree":93,"experienceYears":93,"realName":93,"suggestion":93,"employer":93,"employerType":93,"membershipNumber":93,"other":93,"captcha":93,"modifyTime":93,"page":14,"size":14,"province":93,"regFrom":93,"vipName":93,"timeExpire":93},"2026-09-23T20:57:00",65,"https:\u002F\u002Fapi.speechhome.com\u002FResources\u002FThumb\u002F2026-09-24\u002F2103070353365536768.png","IEEE 国际声学、语音与信号处理会议（ICASSP）是 IEEE 信号处理学会的旗舰会议，每年汇聚全球信号处理、人工智能、通信、感知与多媒体等领域的顶尖专家学者，围绕相关议题展开交流与讨论，推动信号处理理论、方法及应用的创新发展。第52届 ICASSP 将在加拿大多伦多举办，议程包括国际知名学者的…",{"id":185,"title":186,"subjectId":115,"subjectIds":187,"type":19,"author":188,"tag":9,"createTime":189,"likeCount":14,"replyCount":14,"readingCount":173,"coverImage":190,"coverThumbPath":93,"describe":191,"statusCode":14,"sort":14,"subjectEngName":122,"workPlace":9,"salaryRange":9},"2103071658700050432","蚂蚁发布MECT声纹模型：MoE构架刷新VoxCeleb SOTA，并实现100ms流式声纹推理，代码模型已开源","4,2,5",{"id":84,"account":93,"name":86,"level":87,"position":88,"avatar":89,"company":93,"role":93,"introduction":93,"experience":14,"readCount":14,"likeCount":14,"followCount":14,"fansCount":14,"postCount":14,"statusCode":93,"phone":93,"email":93,"follow":93,"createTime":107,"studentID":93,"weChat":93,"hwClassName":93,"traces":93,"jobStatus":93,"country":93,"city":93,"graduatingInstitution":93,"graduationYear":93,"currentInstitution":93,"major":93,"degree":93,"experienceYears":93,"realName":93,"suggestion":93,"employer":93,"employerType":93,"membershipNumber":93,"other":93,"captcha":93,"modifyTime":93,"page":14,"size":14,"province":93,"regFrom":93,"vipName":93,"timeExpire":93},"2026-09-23T19:33:00","https:\u002F\u002Fapi.speechhome.com\u002FResources\u002FThumb\u002F2026-09-24\u002F2103071589976379392.png","近日，蚂蚁集团-大安全机器智能-AI身份智能团队，推出基于混合专家机制（MoE）的声纹识别模型 MECT（Mixture of Experts CNN-Transformer），首次将 MoE 机制引入全监督训练说话人识别领域。在权威基准",{"id":193,"title":194,"subjectId":152,"subjectIds":153,"type":19,"author":195,"tag":9,"createTime":196,"likeCount":14,"replyCount":14,"readingCount":197,"coverImage":198,"coverThumbPath":93,"describe":199,"statusCode":14,"sort":14,"subjectEngName":159,"workPlace":9,"salaryRange":9},"2103056363885301760","ISCSLP 2026丨ASLP@NPU&NJU实验室8篇录用论文分享",{"id":84,"account":93,"name":86,"level":87,"position":88,"avatar":89,"company":93,"role":93,"introduction":93,"experience":14,"readCount":14,"likeCount":14,"followCount":14,"fansCount":14,"postCount":14,"statusCode":93,"phone":93,"email":93,"follow":93,"createTime":107,"studentID":93,"weChat":93,"hwClassName":93,"traces":93,"jobStatus":93,"country":93,"city":93,"graduatingInstitution":93,"graduationYear":93,"currentInstitution":93,"major":93,"degree":93,"experienceYears":93,"realName":93,"suggestion":93,"employer":93,"employerType":93,"membershipNumber":93,"other":93,"captcha":93,"modifyTime":93,"page":14,"size":14,"province":93,"regFrom":93,"vipName":93,"timeExpire":93},"2026-09-23T17:05:00",42,"https:\u002F\u002Fapi.speechhome.com\u002FResources\u002FThumb\u002F2026-09-24\u002F2103056323817115648.png","中文口语语言处理国际学术会议（International Symposium on Chinese Spoken Language Processing, ISCSLP）是国际语音通信协会中文口语语言处理特别兴趣组（ISCA SIG-CSLP）的旗舰会议，每两年举办一次。ISCSLP 2026 将于…",[201,210,212,214,216,218,225,227,229,237],{"id":202,"title":203,"subjectId":93,"subjectIds":93,"type":19,"author":204,"tag":9,"createTime":205,"likeCount":14,"replyCount":14,"readingCount":206,"coverImage":207,"coverThumbPath":93,"describe":208,"statusCode":14,"sort":14,"subjectEngName":209,"workPlace":93,"salaryRange":93},"2108872667896614912","【交我学-43】多模态模型的“思考”方式：视觉、通用音频与语音的 CoT 技术路线",{"id":84,"account":93,"name":86,"level":87,"position":88,"avatar":89,"company":93,"role":93,"introduction":93,"experience":14,"readCount":14,"likeCount":14,"followCount":14,"fansCount":14,"postCount":14,"statusCode":93,"phone":93,"email":93,"follow":93,"createTime":107,"studentID":93,"weChat":93,"hwClassName":93,"traces":93,"jobStatus":93,"country":93,"city":93,"graduatingInstitution":93,"graduationYear":93,"currentInstitution":93,"major":93,"degree":93,"experienceYears":93,"realName":93,"suggestion":93,"employer":93,"employerType":93,"membershipNumber":93,"other":93,"captcha":93,"modifyTime":93,"page":14,"size":14,"province":93,"regFrom":93,"vipName":93,"timeExpire":93},"2026-10-10T18:39:00",85,"https:\u002F\u002Fapi.speechhome.com\u002FResources\u002FThumb\u002F2026-10-10\u002F2108872659323457536.png","大家好，欢迎来到「AudioCC交我学」专栏！ 在文本大模型中，思维链（Chain-of-Thought，CoT）通常意味着模型先写出若干推理步骤，再给出答案。一旦输入变成图像、环境声或连续语音，关键问题就从“是否逐步推理”进一步转向“用什么承载和更新中间思考状态”。 模型可以只在文字空间里推理，也…","audiocc",{"id":113,"title":114,"subjectId":93,"subjectIds":93,"type":19,"author":211,"tag":9,"createTime":118,"likeCount":14,"replyCount":14,"readingCount":119,"coverImage":120,"coverThumbPath":93,"describe":121,"statusCode":14,"sort":14,"subjectEngName":122,"workPlace":93,"salaryRange":93},{"id":84,"account":93,"name":86,"level":87,"position":88,"avatar":89,"company":93,"role":93,"introduction":93,"experience":14,"readCount":14,"likeCount":14,"followCount":14,"fansCount":14,"postCount":14,"statusCode":93,"phone":93,"email":93,"follow":93,"createTime":107,"studentID":93,"weChat":93,"hwClassName":93,"traces":93,"jobStatus":93,"country":93,"city":93,"graduatingInstitution":93,"graduationYear":93,"currentInstitution":93,"major":93,"degree":93,"experienceYears":93,"realName":93,"suggestion":93,"employer":93,"employerType":93,"membershipNumber":93,"other":93,"captcha":93,"modifyTime":93,"page":14,"size":14,"province":93,"regFrom":93,"vipName":93,"timeExpire":93},{"id":124,"title":125,"subjectId":93,"subjectIds":93,"type":19,"author":213,"tag":9,"createTime":128,"likeCount":14,"replyCount":14,"readingCount":129,"coverImage":130,"coverThumbPath":93,"describe":131,"statusCode":14,"sort":14,"subjectEngName":132,"workPlace":93,"salaryRange":93},{"id":84,"account":93,"name":86,"level":87,"position":88,"avatar":89,"company":93,"role":93,"introduction":93,"experience":14,"readCount":14,"likeCount":14,"followCount":14,"fansCount":14,"postCount":14,"statusCode":93,"phone":93,"email":93,"follow":93,"createTime":107,"studentID":93,"weChat":93,"hwClassName":93,"traces":93,"jobStatus":93,"country":93,"city":93,"graduatingInstitution":93,"graduationYear":93,"currentInstitution":93,"major":93,"degree":93,"experienceYears":93,"realName":93,"suggestion":93,"employer":93,"employerType":93,"membershipNumber":93,"other":93,"captcha":93,"modifyTime":93,"page":14,"size":14,"province":93,"regFrom":93,"vipName":93,"timeExpire":93},{"id":134,"title":135,"subjectId":93,"subjectIds":93,"type":29,"author":215,"tag":9,"createTime":137,"likeCount":14,"replyCount":14,"readingCount":138,"coverImage":139,"coverThumbPath":93,"describe":140,"statusCode":14,"sort":14,"subjectEngName":132,"workPlace":93,"salaryRange":93},{"id":84,"account":93,"name":86,"level":87,"position":88,"avatar":89,"company":93,"role":93,"introduction":93,"experience":14,"readCount":14,"likeCount":14,"followCount":14,"fansCount":14,"postCount":14,"statusCode":93,"phone":93,"email":93,"follow":93,"createTime":107,"studentID":93,"weChat":93,"hwClassName":93,"traces":93,"jobStatus":93,"country":93,"city":93,"graduatingInstitution":93,"graduationYear":93,"currentInstitution":93,"major":93,"degree":93,"experienceYears":93,"realName":93,"suggestion":93,"employer":93,"employerType":93,"membershipNumber":93,"other":93,"captcha":93,"modifyTime":93,"page":14,"size":14,"province":93,"regFrom":93,"vipName":93,"timeExpire":93},{"id":142,"title":143,"subjectId":93,"subjectIds":93,"type":19,"author":217,"tag":9,"createTime":145,"likeCount":14,"replyCount":14,"readingCount":146,"coverImage":147,"coverThumbPath":93,"describe":148,"statusCode":14,"sort":14,"subjectEngName":132,"workPlace":93,"salaryRange":93},{"id":84,"account":93,"name":86,"level":87,"position":88,"avatar":89,"company":93,"role":93,"introduction":93,"experience":14,"readCount":14,"likeCount":14,"followCount":14,"fansCount":14,"postCount":14,"statusCode":93,"phone":93,"email":93,"follow":93,"createTime":107,"studentID":93,"weChat":93,"hwClassName":93,"traces":93,"jobStatus":93,"country":93,"city":93,"graduatingInstitution":93,"graduationYear":93,"currentInstitution":93,"major":93,"degree":93,"experienceYears":93,"realName":93,"suggestion":93,"employer":93,"employerType":93,"membershipNumber":93,"other":93,"captcha":93,"modifyTime":93,"page":14,"size":14,"province":93,"regFrom":93,"vipName":93,"timeExpire":93},{"id":219,"title":220,"subjectId":93,"subjectIds":93,"type":19,"author":221,"tag":9,"createTime":222,"likeCount":14,"replyCount":14,"readingCount":181,"coverImage":223,"coverThumbPath":93,"describe":224,"statusCode":14,"sort":14,"subjectEngName":159,"workPlace":93,"salaryRange":93},"2108119561957150720","EvoAudio：让音频大模型在“听力训练”中自进化",{"id":84,"account":93,"name":86,"level":87,"position":88,"avatar":89,"company":93,"role":93,"introduction":93,"experience":14,"readCount":14,"likeCount":14,"followCount":14,"fansCount":14,"postCount":14,"statusCode":93,"phone":93,"email":93,"follow":93,"createTime":107,"studentID":93,"weChat":93,"hwClassName":93,"traces":93,"jobStatus":93,"country":93,"city":93,"graduatingInstitution":93,"graduationYear":93,"currentInstitution":93,"major":93,"degree":93,"experienceYears":93,"realName":93,"suggestion":93,"employer":93,"employerType":93,"membershipNumber":93,"other":93,"captcha":93,"modifyTime":93,"page":14,"size":14,"province":93,"regFrom":93,"vipName":93,"timeExpire":93},"2026-10-05T18:52:00","https:\u002F\u002Fapi.speechhome.com\u002FResources\u002FThumb\u002F2026-10-08\u002F2108119466612232192.png","同一句话播放两遍，内容一模一样，只是第二遍音调更高，让它挑出音调更高的那一遍。能把语音转写准确的模型，未必答得好这样的题。类似的问题还有：一段对话里有几位说话人？一段音乐的节奏有没有加快？这些任务都在提醒我们：知道“说了什么”，不等于听懂声音里的全部信息。 EvoAudio 的思路，是让训练跟着模型…",{"id":150,"title":151,"subjectId":93,"subjectIds":93,"type":19,"author":226,"tag":9,"createTime":155,"likeCount":14,"replyCount":14,"readingCount":156,"coverImage":157,"coverThumbPath":93,"describe":158,"statusCode":14,"sort":14,"subjectEngName":159,"workPlace":93,"salaryRange":93},{"id":84,"account":93,"name":86,"level":87,"position":88,"avatar":89,"company":93,"role":93,"introduction":93,"experience":14,"readCount":14,"likeCount":14,"followCount":14,"fansCount":14,"postCount":14,"statusCode":93,"phone":93,"email":93,"follow":93,"createTime":107,"studentID":93,"weChat":93,"hwClassName":93,"traces":93,"jobStatus":93,"country":93,"city":93,"graduatingInstitution":93,"graduationYear":93,"currentInstitution":93,"major":93,"degree":93,"experienceYears":93,"realName":93,"suggestion":93,"employer":93,"employerType":93,"membershipNumber":93,"other":93,"captcha":93,"modifyTime":93,"page":14,"size":14,"province":93,"regFrom":93,"vipName":93,"timeExpire":93},{"id":161,"title":162,"subjectId":93,"subjectIds":93,"type":29,"author":228,"tag":9,"createTime":164,"likeCount":14,"replyCount":14,"readingCount":165,"coverImage":166,"coverThumbPath":93,"describe":167,"statusCode":14,"sort":14,"subjectEngName":132,"workPlace":93,"salaryRange":93},{"id":84,"account":93,"name":86,"level":87,"position":88,"avatar":89,"company":93,"role":93,"introduction":93,"experience":14,"readCount":14,"likeCount":14,"followCount":14,"fansCount":14,"postCount":14,"statusCode":93,"phone":93,"email":93,"follow":93,"createTime":107,"studentID":93,"weChat":93,"hwClassName":93,"traces":93,"jobStatus":93,"country":93,"city":93,"graduatingInstitution":93,"graduationYear":93,"currentInstitution":93,"major":93,"degree":93,"experienceYears":93,"realName":93,"suggestion":93,"employer":93,"employerType":93,"membershipNumber":93,"other":93,"captcha":93,"modifyTime":93,"page":14,"size":14,"province":93,"regFrom":93,"vipName":93,"timeExpire":93},{"id":230,"title":231,"subjectId":93,"subjectIds":93,"type":24,"author":232,"tag":9,"createTime":233,"likeCount":14,"replyCount":14,"readingCount":234,"coverImage":235,"coverThumbPath":93,"describe":236,"statusCode":14,"sort":14,"subjectEngName":122,"workPlace":93,"salaryRange":93},"2108114572035624960","YODAS v3 开源 110 万小时 48kHz 多通道语音数据，覆盖 147 种语言",{"id":84,"account":93,"name":86,"level":87,"position":88,"avatar":89,"company":93,"role":93,"introduction":93,"experience":14,"readCount":14,"likeCount":14,"followCount":14,"fansCount":14,"postCount":14,"statusCode":93,"phone":93,"email":93,"follow":93,"createTime":107,"studentID":93,"weChat":93,"hwClassName":93,"traces":93,"jobStatus":93,"country":93,"city":93,"graduatingInstitution":93,"graduationYear":93,"currentInstitution":93,"major":93,"degree":93,"experienceYears":93,"realName":93,"suggestion":93,"employer":93,"employerType":93,"membershipNumber":93,"other":93,"captcha":93,"modifyTime":93,"page":14,"size":14,"province":93,"regFrom":93,"vipName":93,"timeExpire":93},"2026-09-29T18:35:00",55,"https:\u002F\u002Fapi.speechhome.com\u002FResources\u002FThumb\u002F2026-10-08\u002F2108114500212363264.jpg","ESPnet 团队发布开放语音数据集 YODAS v3，收录超过 110 万小时、147 种语言的真实世界音频，全部以 48kHz 发布，其中超过 70% 包含两个及以上真正不同的音频通道。 相比此前主要面向语音识别的 YODAS，v3 加入更高带宽、多通道和更丰富的时间戳标注，使同一套数据可以进一…",{"id":169,"title":170,"subjectId":93,"subjectIds":93,"type":29,"author":238,"tag":9,"createTime":172,"likeCount":14,"replyCount":14,"readingCount":173,"coverImage":174,"coverThumbPath":93,"describe":175,"statusCode":14,"sort":14,"subjectEngName":132,"workPlace":93,"salaryRange":93},{"id":84,"account":93,"name":86,"level":87,"position":88,"avatar":89,"company":93,"role":93,"introduction":93,"experience":14,"readCount":14,"likeCount":14,"followCount":14,"fansCount":14,"postCount":14,"statusCode":93,"phone":93,"email":93,"follow":93,"createTime":107,"studentID":93,"weChat":93,"hwClassName":93,"traces":93,"jobStatus":93,"country":93,"city":93,"graduatingInstitution":93,"graduationYear":93,"currentInstitution":93,"major":93,"degree":93,"experienceYears":93,"realName":93,"suggestion":93,"employer":93,"employerType":93,"membershipNumber":93,"other":93,"captcha":93,"modifyTime":93,"page":14,"size":14,"province":93,"regFrom":93,"vipName":93,"timeExpire":93},["Reactive",240],{"$ssite-config":241},{"_priority":242,"env":244,"url":245},{"env":243,"url":14},-15,"production","https:\u002F\u002Fspeechhome.com",["Set"],["ShallowReactive",248],{"post-1630094714327928832":-1,"friend-links":-1,"replies-1630094714327928832":-1,"related-1630094714327928832":-1,"author-posts-1630094714327928832":-1},true,"\u002Fblogs\u002Fnews\u002F1630094714327928832",["Reactive",252],{"user":253,"app":262},{"account":254,"name":256,"userId":257,"avatar":258,"token":259,"role":260,"jobStatus":261},["EmptyRef",255],"\"\"",["EmptyRef",255],["EmptyRef",255],["EmptyRef",255],["EmptyRef",255],["EmptyRef",255],["EmptyRef",255],{"showCompleteProfileModal":263,"theme":265,"mobileMenuOpen":267,"unreadNoticeCount":268},["EmptyRef",264],"false",["Ref",266],"light",["EmptyRef",264],["EmptyRef",92]]</script></body></html>