How to transcribe a YouTube video for free
2026-08-28
YouTube's automatic captions cover over 60 languages, yet they produce a single undifferentiated block of text with no speaker attribution, frequent errors on proper nouns, and accuracy that drops sharply on anything recorded outside a quiet studio. For a 30-minute interview, lecture or panel discussion you want a clean, named transcript - not a wall of auto-generated text that cannot tell the host from the guest. The approach below downloads the audio in one command and turns it into a labelled transcript in minutes, free, with no account needed.
Why you cannot paste a YouTube URL into a transcription tool
Most browser-based transcription services accept a file, not a URL. YouTube does not expose a direct download link in the page source, so there is no way for a web tool to fetch the audio on your behalf without running server-side code that violates YouTube's terms of service. The practical answer is a one-command download on your own machine: you choose what to download, it stays on your drive, and you upload it exactly like any other recording.
This also means the audio travels YouTube → your disk → the transcription service rather than being routed between two third-party servers without your involvement. For recordings of your own content or licensed material that is fine; for anything you do not have the rights to transcribe, that is your responsibility to verify.
Step 1 - Download the audio with yt-dlp
yt-dlp is the standard open-source tool for this task and it runs on Windows, macOS and Linux. Install it once and it works on every video after that.
- macOS (Homebrew):
brew install yt-dlp - Windows: download
yt-dlp.exefrom the project's GitHub releases page and place it somewhere on your PATH. - Linux:
sudo pip3 install yt-dlpor your package manager's equivalent.
Once installed, this single command downloads only the audio track - no video data, so the file is smaller and downloads faster:
yt-dlp -x --audio-format m4a "https://www.youtube.com/watch?v=EXAMPLE"
The -x flag tells yt-dlp to extract audio only. --audio-format m4a
picks a format that WhoSayAh accepts directly; mp3 works just as well if you
prefer it. The file lands in the current directory named after the video's title. A typical
30-minute talk downloads in 20 to 40 seconds on a normal broadband connection.
| Format flag | Output | Typical file size (30 min) |
|---|---|---|
--audio-format m4a | m4a (AAC) | 25-45 MB |
--audio-format mp3 | mp3 | 30-55 MB |
--audio-format opus | opus | 10-20 MB |
If the downloaded file is larger than 100 MB or the recording runs longer than 120 minutes, split it before uploading. Any free audio editor - Audacity on desktop, for example - cuts by time marker without re-encoding. Split at a natural break in the conversation, transcribe the parts separately, then join the text.
Step 2 - Upload and transcribe in WhoSayAh
Drop the downloaded file on the WhoSayAh home page. An upload dialog opens showing the file's name, size and detected duration. No account is needed; the upload starts immediately and both limits are checked before any transcription work begins, so an oversized file is rejected within seconds rather than after a long wait.
- Language. Leave it on Detect automatically unless the recording is short - under 2 minutes - or noisy. For a non-English YouTube video, naming the language explicitly produces cleaner results than auto-detection on shorter clips.
- Speaker labels. For an interview, panel or lecture with audience questions, enable speaker labels and enter the number of distinct voices. For a solo presenter, you can leave labels off; there is nothing to separate.
- Transcribe. Press the button. Lines appear as they finish processing. A 20-minute video typically produces its complete transcript in 3 to 6 minutes.
When the transcript is complete, press Name speakers to replace Speaker 1, Speaker 2 with real names. Each voice is listed alongside how long it spoke and its first line - usually enough to identify who is who without replaying the recording. The names apply across the entire transcript instantly and carry into both download formats.
Why speaker labels matter for YouTube content
YouTube's auto-captions are generated from the audio as a single stream and produce output attributed to no one. For a solo how-to video or a monologue, that is readable. For an interview, a podcast guest episode, a panel discussion or a lecture with audience questions, it is not: you cannot tell who asked what, who made the commitment, or who raised the objection - and neither can a viewer reading the captions.
Speaker labels turn that block into a record. Each line is attributed automatically to the
person who said it. For a 1-hour YouTube interview, that saves 3 to 4 hours of manual
listening and annotation compared with doing it yourself. Once the voices are named, the
.txt reads like a properly attributed conversation, and the .vtt
carries the speaker names in the cue headers - so when you upload it back to YouTube,
viewers can read who is talking.
The interview transcription guide covers the specific settings that matter most when two or more voices are involved - including how to handle the volume imbalance between a host recorded locally and a remote guest, which is one of the most common YouTube recording setups.
Output formats: .txt and .vtt
Once the transcript is ready, you have three options:
- Copy to clipboard - one click, the full labelled transcript lands in your clipboard ready to paste into show notes, a summary document, or a search index.
- .txt download - plain text, one line per speaking turn, speaker name prefixed. Use this for written summaries, a searchable archive of a long video series, or sharing the conversation as a readable document.
- .vtt download - WebVTT with precise timestamps on every cue. YouTube
Studio accepts
.vttdirectly: open the video, choose Subtitles → Add → Upload file, and upload the file. The auto-captions are replaced immediately with the corrected, speaker-labelled version.
Replacing YouTube's auto-captions with a corrected .vtt has a compounding
benefit: caption text is indexed by Google, so a 30-minute interview with accurate captions
becomes thousands of indexable words describing exactly what the video contains. A video
with corrected captions ranks for what was said in it; one without ranks only for its title
and description. For anyone publishing a YouTube channel built around interviews or panels,
that is worth the extra 5 minutes per upload.
The subtitle guide covers the
full attachment workflow in detail - including what to do when a platform expects
.srt instead of .vtt, and how to handle captions that drift
when the file you attach differs from the file you transcribed.
For teams managing a growing catalogue - a course library, a recurring webinar series, a VOD archive updated weekly - the per-file approach scales to dozens of videos before it becomes a bottleneck. At that point the question shifts from transcription to keeping captions synced with an evolving library. CrocOTT handles the platform side of that, keeping captions managed alongside the VOD catalogue rather than attached one file at a time.
What happens to your file
The uploaded audio file is deleted as soon as the transcript is ready. The transcript itself is deleted after 24 hours. There is no account, so nothing links either to you. The privacy policy states exactly what is stored and for how long; the terms of use list the size and length limits in full.
For recordings that cannot leave your own infrastructure - licensed content under a distribution agreement, confidential interviews, anything with data-residency constraints - the alternative is a self-hosted deployment where the audio never reaches an external server. FastoCloud provides the infrastructure for that: processing runs inside your own environment, there is no per-minute charge, and the upload workflow is identical to what is described above.
Frequently asked questions
Does WhoSayAh accept YouTube video files directly?
WhoSayAh accepts uploaded files - m4a, mp3, mp4, webm and others. Download the audio with yt-dlp first, then drop the file on the home page. Files up to 100 MB and 120 minutes are transcribed free with no account.
Can I transcribe a YouTube video in a language other than English?
Yes. Auto-detection works for most languages and handles mixed-language recordings. For short clips or noisy audio in a specific language, name it explicitly in the upload dialog. The output is a transcript in that language - no translation is applied.
Can I add the transcript back to YouTube as captions?
Yes. Download the .vtt file, open YouTube Studio, select the video, go to
Subtitles → Add → Upload file, and upload it. YouTube accepts
.vtt directly and replaces the auto-captions. The corrected, speaker-labelled
captions appear for viewers immediately.
What if the video is longer than the upload limit?
Split the audio into segments under 120 minutes using Audacity or any free audio editor. Cut at a natural pause in the recording, transcribe each segment, and concatenate the transcripts. The file limit is also checked at upload time, so you find out immediately if a segment is still too large.