YouTube's own transcript, and when it is enough
Under many videos, YouTube has a Show transcript button in the description. It opens the captions as a list of lines with their times, scrolls along with the video, and a click on a line jumps to that moment. It is free and needs nothing installed. If you only want to find one moment in one video, use it.
It has limits, and they are the reasons people look further:
- It exists only for videos that have captions. YouTube's help says so directly: the full transcript is there for "any video that has captions". A video without them has no transcript at all.
- Automatic captions come in the video's default language only, so a talk that switches language half way is captioned in one of them.
- It lives on the YouTube page. To keep it, you select and copy by hand, and the lines come out broken the way captions are broken.
A text file with the time of every phrase
VoiceKeyboard does not read YouTube's captions. It downloads the sound of the video and Whisper, a speech recognition model, listens to it on your computer and writes what was said. So it does not matter whether the video has captions or what language the author set as default.
With Pro, turn on "Timestamps in the text" on the FILES screen. Every phrase then begins with the time it starts in the video:
![The FILES screen of VoiceKeyboard with Timestamps in the text ticked. The TRANSCRIPT area shows lines of a lecture, each starting with its time: [00:20:50], [00:20:56], [00:21:01], [00:21:06].](/assets/img/shots/files-batch-en-1920.webp?v=7ed8c10de7)
The result is a plain UTF-8 .txt. That format is useful in a few ways that a web page is not:
- Quoting. Find the sentence with Ctrl+F, and the time is right in front of it. Add it to the YouTube link (for example
&t=12m05s) and the reader lands on the quote. - Notes and summaries. Give the file to ChatGPT, Claude or Gemini and ask for a summary with times; the model can only point to moments that are in the text.
- Archive. A folder of transcripts can be searched with Windows search years later, even if the video is gone.
Without Pro you get the same text without the times. There is no subtitle (.srt) or Word export: the app writes .txt only.
Getting the file
Copy the address of the video, open FILES in VoiceKeyboard, switch to LINK, paste and press START. The first time, the app asks to download two free helper programs (yt-dlp for the download, FFmpeg for the conversion) and a Whisper model; nothing is downloaded until you agree. The whole route, with what you see on the status line, is in turning a YouTube video into text.
Recognition runs on your processor, or on an NVIDIA graphics card if you have one, which is several times faster. You can read the text in the TRANSCRIPT area while the video is still being processed.
When a link does not work
Social networks and video sites such as YouTube, Instagram and Facebook keep fighting the downloading of their content, so this feature is not guaranteed to work: a link that works today may stop working after a change on their side.
Videos that need a sign-in (private, members-only, age-restricted) do not download, because VoiceKeyboard does not use your browser's account. If YouTube refuses a public video, the app checks for a newer version of yt-dlp, which keeps up with changes on YouTube's side, and tries the link once more. If it still fails, the message says so, and the usual fix is to try again later.
Languages
Whisper knows about 99 languages. On Auto, it picks the language from about the first 30 seconds and keeps it for the whole video. If that guess is wrong, choose the language yourself in the LANGUAGE list (twelve are listed there, including English, Spanish, Russian and Japanese). Whisper writes what is said in that language; it does not translate.
Free and Pro
| Free | Pro | |
|---|---|---|
| Video links | 1 a day, up to 60 minutes | no limit, any length, a queue of links |
| Timestamps [00:12:05] | no | yes |
| Recognition | Whisper on your PC | Whisper on your PC |
| Result | .txt | .txt |
Pro costs $14.99 a month, or $9.99 a month billed yearly, and can be refunded within 14 days. A failed or too long attempt does not use up the free link of the day; only a finished transcript counts.
Questions
Can I get timestamps for every word?
No. The time is given per phrase, which is usually a sentence or part of one.
Does it work for a whole playlist?
A link to a video inside a playlist gives that one video. With Pro, paste the links of the videos you need as a list; they are transcribed one after another, each into its own file.
Is the video sent to some transcription service?
No. The sound comes from YouTube to your PC and is recognised there. The downloaded copy is deleted when the job ends; only the .txt stays.
Lectures and courses are the most common case: how to study with lecture transcripts and voice notes shows a way to work with them. For short videos from Instagram, see getting the text of an Instagram Reel.
