[{"data":1,"prerenderedAt":266},["ShallowReactive",2],{"guide-zh-Hans-pull-one-answer-from-a-podcast":3,"guide-zh-Hans-pull-one-answer-from-a-podcast-related":255},{"id":4,"title":5,"author":6,"body":7,"date":235,"description":236,"difficulty":237,"extension":238,"image":239,"lastModified":240,"meta":241,"navigation":242,"outcome":243,"path":244,"readingTime":245,"robots":239,"scenario":246,"seo":247,"stem":249,"tools":250,"__hash__":254},"guides_zh_Hans\u002Fzh-Hans\u002F8.guides\u002F6.pull-one-answer-from-a-podcast.md","播客转文字：两小时的节目只转录你需要的那一段","FlowPick Team",{"type":8,"value":9,"toc":224},"minimark",[10,14,17,22,25,28,31,35,47,56,69,76,82,85,89,92,100,103,109,114,117,123,128,132,140,143,146,152,157,161,164,180,183,186,193,199,205,211,214,217],[11,12,13],"p",{},"某人在播客里说了句有用的话。你大概记得位置——前三分之一，第二个广告之后。你需要准确的原话，用来引用、做笔记或者回复别人。",[11,15,16],{},"最直觉的做法是把整期转录成文字再搜。这确实可行，但也是最慢的一条路。这篇讲快的那条。",[18,19,21],"h2",{"id":20},"为什么整期全转录是错的默认选项","为什么\"整期全转录\"是错的默认选项",[11,23,24],{},"识别成本跟时长成正比。两小时的音频大约是半小时节目的四倍时间，产出的文稿里有九成跟你需要的那句无关。",[11,26,27],{},"播客本身的结构还有利于先剪：广告段、自我推广、重复的回顾，这些本来就是你听的时候要跳过的部分，却会原样进入转录。",[11,29,30],{},"正确的顺序是：一边快放一边定位、把那段剪出来、只转录那几分钟。",[18,32,34],{"id":33},"第一步拿到播客的音频文件","第一步：拿到播客的音频文件",[11,36,37,38,42,43,46],{},"找到音频源。多数播客页面上的音频就是一条 MP3 或 M4A 直链，确认的办法是在 Network 面板筛 ",[39,40,41],"code",{},"media","，或者看页面源码里的 ",[39,44,45],{},"\u003Caudio>"," 元素、RSS 里的 enclosure 地址。",[11,48,49,50,55],{},"把链接粘进",[51,52,54],"a",{"href":53},"\u002Fzh-Hans\u002Faudio-downloader","音频下载器","。它接受 MP3、M4A\u002FAAC、WAV、OGG、FLAC 的直链。",[11,57,58,59,63,64,68],{},"如果这档播客给的是 HLS 音频清单而不是直链——平台托管的节目越来越常这样——改用 ",[51,60,62],{"href":61},"\u002Fzh-Hans\u002Fm3u8-downloader","M3U8 下载器","，读清单的做法见",[51,65,67],{"href":66},"\u002Fzh-Hans\u002Fguides\u002Farchive-a-course-over-hls","课程归档那篇","。",[11,70,71],{},[72,73],"img",{"alt":74,"src":75},"音频直链与待保存 WAV 结果，文件体积为 180.5 KB","\u002Fimg\u002Fguides\u002Faudio-downloader-url.jpg",[11,77,78],{},[79,80,81],"em",{},"英文界面，直接获取本地公开地址上的生成语音，无需扩展。结果面板显示格式和体积；检查时长需点击 Inspect audio（检查音频）。",[11,83,84],{},"动手之前确认一件事：文件时长对不对。两小时的节目下来一个 12 分钟的文件，那是预告片，不是正片。",[18,86,88],{"id":87},"第二步定位然后剪出来","第二步：定位，然后剪出来",[11,90,91],{},"这是最多人跳过的一步，也是最能省时间的一步。",[11,93,94,95,99],{},"用",[51,96,98],{"href":97},"\u002Fzh-Hans\u002Ftools\u002Faudio-trimmer","音频裁剪","打开下载好的文件。能快放就用 1.5 倍或 2 倍听，或者直接看波形拖，记下入点和出点。目标是一个两到十分钟的窗口，里面装着完整的论述——不是只要那一句。",[11,101,102],{},"故意剪宽一点是有原因的。三句话的文稿根本看不出这句话在回答什么，你最后还是得回去听。",[11,104,105],{},[72,106],{"alt":107,"src":108},"两小时的播客剪成一个八分钟的窗口，只转录这一段而不是整期","\u002Fimg\u002Fguides\u002Fpodcast-excerpt.svg",[11,110,111],{},[79,112,113],{},"先定位，再剪，最后转录。顺序对了，这件事是几分钟，不是一小时。",[11,115,116],{},"顺手把前后文里的口播广告剪掉。那些内容没有信息量，而且是识别效果最差的部分——人声底下压着音乐和音效。",[11,118,119],{},[72,120],{"alt":121,"src":122},"音频波形中的三秒选区，以及可编辑的起止时间","\u002Fimg\u002Fguides\u002Faudio-trimmer-range.jpg",[11,124,125],{},[79,126,127],{},"英文界面，使用生成语音演示，并非播客录音。在自己的节目中用同样的起止时间控件选取需要的八分钟片段。",[18,129,131],{"id":130},"第三步转录这一段","第三步：转录这一段",[11,133,134,135,139],{},"把剪出来的片段交给",[51,136,138],{"href":137},"\u002Fzh-Hans\u002Ftools\u002Faudio-to-text","音频转文字","。播客音频通常是最容易处理的一类素材：一到两个人、像样的麦克风、语言稳定。",[11,141,142],{},"语言要自己指定，别依赖自动检测。检测是取开头几秒来判断的，而一段从句子中间开始的片段，可用的判断依据比整期少。",[11,144,145],{},"段落里有人名、数字或专业术语，选标准模型——需要引用的一段里通常都有。轻量模型够用来确认\"对，就是这个部分\"。",[11,147,148],{},[72,149],{"alt":150,"src":151},"音频转录结果，显示时间戳、可编辑文本和字幕导出控件","\u002Fimg\u002Fguides\u002Faudio-to-text-clip.jpg",[11,153,154],{},[79,155,156],{},"英文界面，Whisper Tiny 在 CPU 上实际识别生成的英语语音。这里只演示界面，不代表播客识别质量。",[18,158,160],{"id":159},"第四步引用之前先核对","第四步：引用之前先核对",[11,162,163],{},"识别能抓到大概意思，但漏掉的正好是引用时最要紧的细节。把一句话归到某个人名下之前：",[165,166,167,171,174,177],"ul",{},[168,169,170],"li",{},"专有名词回到那个时间点重听一遍",[168,172,173],{},"数字重点核对——\"十五\"和\"五十\"是已知的易错点",[168,175,176],{},"确认是谁说的。这个版本不区分说话人，两人对谈时你得自己对上是哪一位",[168,178,179],{},"笔记里留下时间点。万一这个说法有争议，节目里的时间戳就是你的依据",[11,181,182],{},"文稿里的时间点是相对于你剪出来的片段，不是相对于整期节目。要在整期里标时间，记得加上第二步的偏移量。",[18,184,185],{"id":185},"中途出问题怎么办",[11,187,188,192],{},[189,190,191],"strong",{},"下载链接 403。"," 播客 CDN 会给地址签名，复制来的链接会过期。重新打开节目页复制一条新的。",[11,194,195,198],{},[189,196,197],{},"片段从半个字开始。"," 转录开头是个碎片，模型容易读错。入点比你以为的再多剪两秒。",[11,200,201,204],{},[189,202,203],{},"文字都对，但顺序乱了。"," 长停顿和抢话会干扰分段。短片段一般手工调几下就好。",[11,206,207,210],{},[189,208,209],{},"这期节目要登录才能听。"," 这套流程的前提是音频公开可访问。它不是付费墙的绕法，工具也没有往那个方向做。",[18,212,213],{"id":213},"收尾",[11,215,216],{},"一段八分钟的音频、它的文稿，两者都有时间点，再记一段\"这个片段在整期里的偏移量\"。音频以后还能再听一遍核对，光有文稿做不到。",[11,218,219,220,68],{},"播客和音频场景的更多做法，见",[51,221,223],{"href":222},"\u002Fzh-Hans\u002Fdocs\u002Fusecases\u002Fpodcasts-audio","播客与音频下载",{"title":225,"searchDepth":226,"depth":226,"links":227},"",2,[228,229,230,231,232,233,234],{"id":20,"depth":226,"text":21},{"id":33,"depth":226,"text":34},{"id":87,"depth":226,"text":88},{"id":130,"depth":226,"text":131},{"id":159,"depth":226,"text":160},{"id":185,"depth":226,"text":185},{"id":213,"depth":226,"text":213},"2026-10-29","为了引用一句话去转录两小时的播客，是最慢的一条路。先下载音频、快放定位、剪出那几分钟，再只转录剪出来的片段。从播客链接到可引用文稿的最短路径。","basic","md",null,"2026-11-01",{},true,"一段八分钟的音频和它的文稿，而不是两小时的。","\u002Fzh-hans\u002Fguides\u002Fpull-one-answer-from-a-podcast",6,"transcribe",{"description":248,"title":5},"播客音频转文字的方法：想引用某句话时，先用音频下载器拿到 MP3\u002FM4A 直链文件（HLS 清单则用 M3U8 下载器），快放定位后剪出两到十分钟的片段，只转录这一段，省下整期转录的等待。附下载链接 403 过期、片段开头识别出错等问题的处理，以及引用前的核对清单。","zh-Hans\u002F8.guides\u002F6.pull-one-answer-from-a-podcast",[251,252,253],"\u002Faudio-downloader","\u002Ftools\u002Faudio-trimmer","\u002Ftools\u002Faudio-to-text","s-wM1YuBNjeJkYmr08nnleSgTDMYVloXf4GcoAdBxHM",[256,264],{"path":257,"title":258,"description":259,"tools":260},"\u002Fzh-hans\u002Fguides\u002Fturn-a-meeting-recording-into-notes","会议录音转文字：把会议视频变成可搜索的纪要文稿","会议录屏先剪掉废话、必要时转成 MP4，再在本地转录成可编辑的文稿和 SRT 字幕。模型怎么选、语言怎么设、人名数字怎么校对，以及长录音中断后怎么办。",[261,262,263],"\u002Ftools\u002Fvideo-trimmer","\u002Ftools\u002Fwebm-to-mp4","\u002Ftools\u002Fvideo-to-text",{"path":244,"title":5,"description":236,"tools":265},[251,252,253],1793546746385]