youtubegptTranscript generator › Translated subtitles
YouTube subtitles in another language
Many videos already carry caption tracks in a dozen languages or more. Extract the one you want directly — and when it does not exist, translate from clean prose rather than from fragments.
Check what already exists first
People reach for a translator before checking whether the video already has the language they need. It often does. In our testing the track counts ranged widely: a music video carried 6, a maths lecture carried 31, a Korean pop video carried just 1.
After extraction, every track is listed with its language and whether it is human-authored or machine-made. If yours is there, take it directly — a caption track written or reviewed by a human beats anything you would produce by translating afterwards.
Via the API, request format=json and read availableTracks, then re-request with lang set:
curl "https://youtubegpt.ai/api/transcript?v=VIDEO_ID&lang=es&format=paragraphs"Three kinds of track, and they are not equal
| Kind | How to spot it | Quality |
|---|---|---|
| Human-authored, original language | Marked as not auto-generated, matches the spoken language | Best. Usually punctuated and proofread. |
| Human-authored translation | Not auto-generated, different language from the speech | Good. Added by the creator or a contributor. |
| Auto-generated / auto-translated | Flagged as auto-generated | Variable. Machine transcription, sometimes machine translation on top of it. |
Auto-translated tracks are a translation of a transcription
When YouTube offers a language the creator never provided, it is usually machine-translating its own machine transcription. Two lossy steps stacked: recognition errors on names and jargon get carried into the translation and can come out as confident nonsense. Prefer a human track when one exists, and treat auto-translated output as a rough guide.
When the language you need is not there
Translate it yourself — but translate the right thing. This is where most people lose quality without realising.
Do not translate raw captions
Caption fragments are cut on timing boundaries, not sentence boundaries, so a single sentence is often split across three lines. Feed those fragments to a translator and it translates each stub without the rest of the sentence for context. Grammatical gender, verb tense and word order all depend on context the translator cannot see, and the result reads like it was assembled from parts — because it was.
Translate the readable version
Use the Readable tab, where fragments have been rejoined into paragraphs. Whole sentences give the translator the context it needs, and the output is markedly better. Then paste into whichever translator or language model you use:
Translate the following transcript into Spanish. Keep the
speaker's register — do not formalise casual speech. Leave
proper nouns and technical terms in the original language
where there is no established translation.That last instruction matters more than it sounds. Translators love to "translate" product names and technical terms into something that does not exist.
If you need timed subtitles back
Translating paragraphs breaks the line-to-timecode mapping, so you cannot simply paste the result into an SRT. Download the SRT and translate the text lines while leaving the numbering and timecodes untouched — awkward but it preserves timing. Expect to adjust line lengths afterwards, since translations frequently run longer than the original and overflow the subtitle box.
Reading foreign-language videos
A common use that has nothing to do with subtitling: extract the original-language transcript, translate the prose, and read it. For a lecture or interview this is often faster and more useful than watching with subtitles — you can skim, search and re-read, none of which the player allows.
It works particularly well combined with summarising: ask a model to summarise a foreign-language transcript directly in your language, and you get the gist of a 40-minute talk in a minute or two.
Frequently asked
How do I see which languages a video has?
Extract it — every available track is listed with its language and whether it is auto-generated. Via the API, read availableTracks from the JSON response.
Can I get a transcript in a language the video does not offer?
Not directly; only tracks that exist can be extracted. Take the readable transcript and translate it yourself, which also gives you better output than an auto-translated track.
Why is the auto-translated version poor?
Because it is usually a machine translation of a machine transcription. Recognition errors in the first step get faithfully translated in the second.
Can I translate an SRT and keep the timing?
Yes, if you translate only the text lines and leave the cue numbers and timecodes untouched. Check line lengths afterwards — translations often run longer and overflow.
Which languages are supported?
Whatever the video carries. There is no fixed list on this side; the tracks come from the video itself.
More guides
Independent tool. youtubegpt.ai is not affiliated with, endorsed by, or operated by YouTube or Google. It works with publicly available caption data from YouTube and does not host, mirror or re-upload any video. Transcript text belongs to the original video creators. "YouTube" is a trademark of Google LLC.