Home首页 / Guide使用说明
Guide使用说明
From install to a finished transcript takes a few minutes — the installer is large because every speech model is inside it, so there is nothing to download afterwards.从安装到拿到成品文字只要几分钟 —— 安装包大是因为语音模型全在里面，装完之后不需要再下载任何东西。
1 · Install一、安装
- Download the installer from the home page and run it. It installs for the current user only — no admin rights, no services.从首页下载安装包并运行。只装给当前用户，不需要管理员权限，不装服务。
- On first launch AV2Text checks the bundled models and starts working — no download step. If a model file is ever missing or added later, it fetches just that file.首次启动只做一次模型自检，然后就能直接干活 —— 没有下载步骤。将来若某个模型缺失或新增，也只会补下那一个文件。
- After that it works with the network off.下完之后，断网也能用。
2 · Three ways to transcribe二、三种转写方式
Drop in a file拖入文件
Drag an audio or video file onto the window, or use the folder button in the toolbar. The task appears in the queue on the left with its progress.把音视频文件拖到窗口上，或者用工具条上的文件夹按钮选文件。任务会出现在左侧队列里，带进度。
Paste a link粘贴链接
Paste a video or audio page URL into the box in the middle of the toolbar and press the link button. AV2Text downloads just the audio track and then transcribes it.把视频或音频页面链接粘到工具条中间的输入框，点旁边的链接按钮。AV2Text 只取音轨，取完再转写。
If a site refuses: some sites answer a bogus "412 / timed out" to automated requests. The app names the site and says what happened; waiting a bit or trying another link usually works.如果站点拒绝：有些站点会对自动请求回一个假的「412 / 超时」。程序会点名是哪个站点、发生了什么；等一会儿或换个链接通常就好了。
Live from the mic麦克风实时
Press the mic button and speak. Text lands on screen less than a second behind. Allow microphone access the first time Windows asks.点麦克风按钮开始说话，文字不到一秒就上屏。第一次 Windows 问麦克风权限时允许即可。
3 · The three panes三、三页视图
Original, Corrected and Translated are three views of one transcript, not three separate documents. They share the same blocks and the same timecodes.原文、校正、译文是同一份内容的三张脸，不是三份各自的文件。它们共用同一套分块与同一套时间码。
- Click into a block and type — the block becomes editable in place.点进某个块直接改字，改的就是那个块本身。
- What you change applies to all three panes: block structure, speaker names, speaker labels, selection order.改动会作用到三页：块结构、说话人名字、说话人标签、选中序号。
- Renaming a speaker once renames them everywhere.说话人改名一次，三页都跟着改。
4 · Export四、导出
| Format格式 | Use it for什么时候用 |
|---|---|
| TXT | Reading, pasting into documents阅读、贴进文档 |
| SRT | Video subtitles (timecodes included)视频字幕（带时间码） |
| LRC | Music players and lyric files播放器歌词 |
| DOCX | Handing a document to someone把文档交出去 |
| Reading or archiving without editing不必再改的阅读或存档 | |
| MD | Notes and docs that keep some formatting要保留一点格式的笔记与文档 |
| JSON | Feeding the transcript into another program把转写结果喂给别的程序 |
Pick the export format and the save folder in the toolbar ("save path").导出格式和保存目录都在工具条上（「保存路径」）。
5 · When something goes wrong五、出问题了怎么办
| Symptom现象 | What to do怎么办 |
|---|---|
| The engine does not answer引擎没反应 | Close and reopen the app. If it keeps happening, use Settings → About & support → export a diagnostics file.关掉重开。反复出现就用 设置 → 关于与支持 → 导出诊断包。 |
| Model download stalls模型下载卡住 | It retries on its own and can resume. A slow network is usually the cause, not a broken file.它会自己重试并支持断点续传。通常是网络慢，不是文件坏了。 |
| No audio from the mic麦克风没声 | Windows settings → Privacy → Microphone → let desktop apps use the microphone.Windows 设置 → 隐私和安全性 → 麦克风 → 允许桌面应用访问。 |
| A long file freezes长文件转写卡死 | Memory ran out. Close other heavy apps and keep about 4 GB free.内存不够。关掉其它吃内存的程序，留出约 4 GB。 |
6 · What goes online六、什么时候会联网
- Transcription: never. It runs on your CPU.转写：不联网，跑在本机 CPU 上。
- Model download is only used for a missing or newly added model — a normal install never downloads.模型下载只在「文件缺失或新增模型」时才用；正常安装不会下载任何东西。
- Pasting a link: fetches that page's audio track.粘贴链接：去那个页面取音轨。
- Cloud translation, if you turn it on: text only. See the privacy policy.云端翻译（你主动开启才用）：只发文字。见隐私政策。