Here is the failure you keep hitting: you paste a YouTube URL into a transcript tool and get an empty result, or “no captions found.” The video is perfectly fine — it just has no subtitle track. And because nearly every “YouTube transcript” tool is really a caption extractor, a missing track is a hard stop.
This page is for that exact case. Instead of giving up, the transcriber downloads the video's audio and runs Whisper-class speech recognition on it, building a fresh, timestamped transcript from the sound itself.
“Extract the captions” is not the same as “transcribe”
The word “transcript” hides two completely different operations:
- Extraction reads the caption track that already exists on the video. It is instant and exact — but it only works when a track is there.
- Transcription listens to the audio and writes the text from scratch. It works whether or not a caption track exists. It is what happens when a video has no subtitles.
A tool that only extracts is the common kind — and it is the kind that fails on the video you are looking at. The tool on this site does both: it prefers extraction when captions exist, and falls back to transcription when they do not.
How the no-subtitle path works
- Paste the YouTube URL. If a caption track exists you get the result immediately.
- If not, the tool starts an AI transcription job and returns a job ID in a second or two.
- The job runs on its own hardware with a Whisper-class model. When it finishes, the transcript appears with the usual export buttons.
What you get
- A timestamped transcript — same shape as a caption extraction, per-line timing included.
- TXT (timestamped lines), SRT and VTT subtitle files, and JSON (start, duration, text) for pipelines.
- Translation of the result into another language, and bilingual SRT if that is what you need.
When you hit this
- Podcasts, talks, and lectures uploaded without captions.
- Videos whose captions were removed or are only in one language.
- Older uploads, or creators who simply never turned captions on.
- Any video you want as a clean, citable text document.
Related
The feature behind this page is the YouTube Transcriber. If your content is full of jargon that speech models mangle, see transcription with a custom word list. For the whole product — batch, translation, summaries, and the API — start with the free tool.
Frequently asked questions
Why can't most YouTube transcript tools handle a video without subtitles?
Because almost all of them extract the caption track that already exists on the video — they do not transcribe the audio. If there is no caption track (or it has been removed, or it exists only in a language you do not have access to), there is nothing to extract, so the tool fails. A tool that can transcribe the audio is a different thing entirely, and it is exactly what this page is about.
How does AI transcription of a YouTube video work here?
The video's audio track is downloaded and passed to a Whisper-class speech recognition model, which produces a brand-new transcript with per-segment start times. It runs as an asynchronous job: you get a job ID within a couple of seconds and the result as soon as the audio is processed. The output is the same shape as a caption extraction — segments with start time, duration, and text — so everything downstream (copy, SRT, VTT, JSON, translation) works identically.
Does it automatically decide whether to use captions or AI?
Yes. By default the tool prefers the caption track when one exists, because that is the exact source text and it is instant and cheap. Only when captions are missing — or when you explicitly force AI — does it fall back to transcribing the audio. You get a transcript either way without having to think about which path it took.
How accurate is it, and how long does it take?
For clear speech in common languages, Whisper-class models reach high-90s word accuracy; heavy accents, music, or multiple overlapping speakers reduce it. A typical 5–10 minute video transcribes in a few minutes. ASR is billed per second of audio (1 credit per 90 s), not per video, so a long lecture costs proportionally more.
What can I do with the transcript once it is generated?
Everything you can with a caption-based one: copy it, download TXT with timestamps, export SRT or VTT subtitle files, or pull the machine-readable JSON into a pipeline. You can also translate the result into another language and, if the audio is technical, bias it with a custom word list on the local-file upload.
When would I want to force AI even if captions exist?
When the existing captions are low quality (misspelled auto-captions, wrong language, or a sparse track) and the audio is clean. Forcing the transcription path regenerates the text from the audio, which often beats a bad caption track. The trade-off is the ASR cost and a short wait instead of an instant caption fetch.