What you need
- VoiceKeyboard on Windows 10 or 11, and the internet while the sound downloads.
- A Whisper model, downloaded once inside the app. The recommended size, Best · Large v3 Turbo, is 547 MB.
- Two free helper programs: yt-dlp (about 17 MB) fetches the sound behind the link, FFmpeg (about 73 MB) converts it for Whisper. The first time, a window lists the size, source and licence of each and downloads nothing until you press Download.
From link to text in seven steps
- Open the video in your browser and copy the address of its page.
- In VoiceKeyboard, press FILES in the top bar and choose LINK under SOURCE & OUTPUT.
- Press Paste, or drag the link from the browser onto the FILES screen.
- Under TRANSCRIPT DESTINATION, press Save as... and choose the folder and the name of the .txt file.
- Under LOCAL MODEL, press "Choose model…" if no model is set yet, and download one.
- Leave LANGUAGE on Auto, or pick the language of the video if you know it.
- Press ▶ START.
The status line first reads DOWNLOADING… with the percentage, size and speed. Then it switches to DECODING… with its own percentage and the name of what does the work: CUDA for an NVIDIA graphics card, BLAS or CPU for the processor. The text appears in the TRANSCRIPT area phrase by phrase, so you can start reading long before the end. When the video is done, the .txt is saved.
What happens to the video
VoiceKeyboard asks the site for the sound track only, where the site offers one, and keeps nothing of the video. FFmpeg turns the sound into the plain format Whisper reads, and Whisper writes the text on your processor or graphics card. The working copies sit in the Windows temporary folder and are deleted when the job ends, whether it finished, failed or was cancelled. Your .txt is all that stays.
The only server involved is the video site itself, which sends the sound. The audio is not passed on to any transcription service.
Timestamps, copying and the text file
With Pro, tick "Timestamps in the text" and every phrase starts with its time in the video, like [00:12:05]. That is the quickest way back to the moment a quote was said.
The result is a plain UTF-8 .txt, and Copy puts the whole text on the clipboard for Word, Google Docs or a chat with an AI that should summarise it. There is no subtitle (.srt) or Word export. A transcript never replaces an existing file: if the name is taken, the new one gets (2) in its name.
Several links in one go
Pro takes a whole list. Paste the links one per line or separated by spaces or commas; the box then shows the first one and "+N more". They download one at a time and run one after another, each with its own transcript and its own file, so you can start a series of lectures and come back later. On the free plan, only the first link of a pasted list is kept.
Which sites work
VoiceKeyboard hands the link to yt-dlp, a free open-source downloader that handles YouTube and many other video and audio sites. If yt-dlp can get the sound of a page, FILES can transcribe it. Some links still fail:
- Videos that need you to be signed in (private, members-only or age-restricted) do not download, because the app does not use your browser's login.
- A link to a video inside a playlist gives that one video, not the whole list.
- Video sites change, and the copy of yt-dlp the app uses does not update itself. When a link that plays fine in the browser ends with ERR: DOWNLOAD_FAILED, a change on the site is the likely reason.
FILES or an online engine
VoiceKeyboard has two online engines, Google and Soniox, and they work on the LIVE screen: they type what your microphone hears, while it is said. FILES has no online mode. Every link and every file is recognised by Whisper on your computer.
| FILES with a link | LIVE with Google or Soniox | |
|---|---|---|
| Source | a video that already exists | speech into the microphone, right now |
| Recognised by | Whisper, on your PC | Google's or Soniox's servers |
| Internet | for the download only | the whole time |
| Speed | depends on your PC; several times faster with an NVIDIA card | text appears as you speak |
| Result | a .txt file, with timestamps on Pro | text typed into the window you work in |
Pick LIVE with an online engine when the words are being spoken into your microphone now and you want them typed at once (Google is free, Soniox is part of Pro; voice typing in English compares the engines). Pick FILES when the talk is already recorded. On an older laptop without an NVIDIA card a long video takes a while; the smaller Whisper sizes are faster but make more mistakes.
Free and Pro for links
Free gives one link a day, up to 60 minutes, without timestamps, plus one audio or video file from your disk. The length is known only after the download, so a longer video stops with FREE: UP TO 60 MIN. That attempt does not use up the day's link, and neither does a failed one: only a finished transcript counts, and the allowance starts again at midnight.
Pro removes the limits: any length, a queue of links, timestamps. It costs $14.99 a month, or $9.99 a month billed yearly, with a 14-day refund. The table of both plans is in the guide to transcribing audio files.
Questions about video links
Does the recognition need the internet?
No. The internet is used to download the sound of the video and, the first time, the Whisper model and the two helper programs. Whisper itself works offline.
Which languages can the video be in?
Whisper knows about 99 languages. On Auto it decides from roughly the first 30 seconds and keeps that language for the whole video, so parts in a second language may come out wrong. If the guess is wrong, the LANGUAGE list fixes it to one of twelve: Russian, English, Ukrainian, German, Spanish, French, Italian, Portuguese, Polish, Turkish, Japanese or Chinese.
Can it translate the video into English?
No. Whisper writes down what is said, in the language it is said in.
Can I keep dictating while a video is transcribed?
Yes. FILES and dictation run side by side, and FILES does not use the 2,000 free dictation words of the day.
Recordings already on your computer, such as a meeting or a voice memo, go through the same screen: see transcribing audio and video files offline. Every button of FILES is described in the help page Transcribing files and video links.
