shrybe
need captions? drop the URL – I'll write down every line, timed.

Captions for any
video.

Start from a YouTube link, podcast episode, or audio URL. shrybe writes a timestamped transcript and exports it straight to SRT or VTT subtitles. Free, no Chrome extension, no watermarks.

· audio or video, up to 100 MB

auto often misreads Slavic — pick if it gets it wrong
Timestampsstart time on every paragraph
WCAG-friendlyreview once, ship accessible
One‑tap from iOS Share any link to shrybe, get a transcript in 30 sec →

Captions are the lowest-effort accessibility win on the internet

The World Health Organization words it precisely, and the precision matters: over 5% of the world's population – 430 million people – require rehabilitation to address disabling hearing loss. By 2050 the WHO projects that figure will pass 700 million – one person in ten – with nearly 2.5 billion living with some degree of hearing loss. Captions are not a niche feature. They are also how most social video is consumed – autoplay-on-mute is the default behaviour on Twitter/X, Instagram, TikTok, and LinkedIn. A video without captions is, for the typical scrolling viewer, a video without sound.

And captions are the cheapest possible accessibility intervention. There is no UI redesign, no screen-reader audit, no developer ticket. The audio you already recorded, rendered as text. shrybe is the part where you get the text.

Three steps to a subtitle file

  1. Transcribe the source. Paste your YouTube / podcast / audio URL. Whisper writes a timestamped transcript – one paragraph per audio segment, with the start time stamped on the left.
  2. Open the Download menu. Every transcript exports as .txt, .srt, .vtt, or .json – the subtitle files are built from Whisper's own segment timestamps, nothing to convert.
  3. Pick .srt or .vtt. SRT for YouTube and almost every platform; VTT for HTML5 <track> elements. The format spec below is for hand-editing afterwards.

The SRT and VTT formats, in 30 seconds

SRT is older and simpler. It looks like this: 

1
00:00:00,000 --> 00:00:04,200
Welcome back to the show.

2
00:00:04,200 --> 00:00:07,800
Today we are talking about transcription.

VTT is the W3C-standard cousin used by HTML5 <track> elements: 

WEBVTT

00:00:00.000 --> 00:00:04.200
Welcome back to the show.

00:00:04.200 --> 00:00:07.800
Today we are talking about transcription.

The differences are small and exact – worth getting right the moment you hand-edit a file:

For upload to YouTube, Vimeo, Loom, or most CMS video players, save the file as .srt. For HTML5 video directly in a browser, save as .vtt. Most editors and players accept either.

The WCAG bar and where AI captions fit

WCAG 2.2 Success Criterion 1.2.2 requires captions for prerecorded audio in synchronised media at the Level A conformance bar. It does not require captions to be human-authored – it requires them to be accurate. Auto-generated YouTube captions famously fail this test on accented speech, jargon, and proper nouns. A reviewed Whisper transcript easily clears it.

Two criteria are worth keeping apart, because they are not the same bar. 1.2.2 Captions (Prerecorded) sits at Level A – the baseline, and the one a recorded video has to meet. 1.2.4 Captions (Live) sits at Level AA and covers live audio in synchronised media. A file-based tool like this one cannot help you with the second: there is no recording to paste yet. We would rather name the gap than let the two blur into "WCAG-compliant".

The realistic workflow for a creator: generate the first pass with shrybe, open the .txt in any editor, fix the names and the brand terms (Whisper consistently mangles rare-in-training words), then export as SRT/VTT. Twenty minutes of editing on an hour of video. That is the difference between auto-captions and good captions.

What Whisper gets right and wrong

Per the Whisper paper (Radford et al., 2022), English word error rate across the fourteen English datasets in Table 2 runs from 2.7% on clean read speech to 36.4% on far-field meeting audio, averaging 12.8%. The spread is driven by how the audio was captured and what kind of speech it is – the same AMI meetings score 16.9% on a head-worn microphone and 36.4% on a table one. Accent matters too, and the paper does not hide it: the dialect-focused CORAAL corpus sits at 16.2%. Across languages, OpenAI's own model card is equally blunt – the models "perform unevenly", with lower accuracy on low-resource ones. Caveats: technical jargon and proper nouns drift. Brand names that did not appear in Whisper's training corpus become approximations. Overlapping speech becomes a single garbled paragraph because Whisper does not separate voices.

The honest workflow: AI for the first draft, human eyes for the last 5%. Most teams find that ratio cheaper than a fully-human pass and substantially better than fully-auto.

Frequently asked

Why do captions matter – beyond accessibility?

Three reasons that compound. First, accessibility: the WHO counts 430 million people who require rehabilitation for disabling hearing loss, and projects that past 700 million – one person in ten – by 2050. Second, comprehension in noisy environments: a large share of video on social platforms is watched on mute, so captions are how most viewers actually receive the content. Third, search: caption files are indexed by YouTube and most video platforms, which means accurate captions help discovery.

Does shrybe export SRT or VTT files directly?

Yes – free. The Download menu on every transcript exports .srt and .vtt (plus .txt and .json) directly, with timestamps straight from Whisper's segments. The format spec below is for when you want to edit or debug a subtitle file by hand.

What is the difference between SRT and VTT?

Both are timestamped subtitle file formats. SRT (.srt) is older and works with nearly every video platform, player, and editing tool. WebVTT (.vtt) is the W3C standard for HTML5 <track> elements and supports cue positioning, styling, and metadata that SRT doesn't. If you don't know which to pick: SRT for upload to YouTube/Vimeo, VTT for embedding directly into an HTML5 video tag.

How accurate are AI-generated captions compared to human captions?

Whisper-grade AI captions are good enough for accessibility on clear English audio, and the source data is specific about what "clear" is worth: across the fourteen English datasets in Radford et al. (2022, Table 2), Whisper Large V2 spans 2.7% word error on clean read speech to 36.4% on far-field meeting audio, averaging 12.8%. Human captioning still wins on dense technical content, multi-speaker live events, and any context where one wrong word changes meaning. A common workflow: generate the first pass with AI, then a human editor cleans up names, jargon, and overlapping speech.

Are AI captions WCAG-compliant?

AI captions can meet WCAG 2.2 Success Criterion 1.2.2 (Captions for prerecorded media) provided they are accurate. Inaccurate captions – including auto-generated YouTube captions with obvious errors – do not pass WCAG. The rule is that captions must convey the audio content equivalently, not that they must be human-created. A reviewed Whisper transcript easily meets that bar; an unreviewed one is a coin flip.

Can I generate captions for non-English videos?

Yes. Whisper covers 100 languages, and the language picker lets you choose the source language explicitly. For Ukrainian Surzhik, Belarusian, and other low-resource Slavic content, our metadata pre-detection helps Whisper choose the right model setting before transcription starts. Expect highest accuracy on English, Spanish, French, German, Mandarin, Japanese, and Portuguese – those have the most Whisper training data.

Should I caption my own videos before YouTube auto-captions them?

Yes, if accuracy or rankings matter. YouTube's auto-captions have improved dramatically over the past few years but still misread names, jargon, brand terms, and accented speech. Uploading your own caption file overrides the auto-captions and is indexed for search. The hour you spend cleaning up a generated caption file is the most direct accessibility-and-SEO win available.

Do captions help with search engine rankings?

Indirectly. Google does not officially weight captions as a ranking factor, but the text in your caption file is indexed and surfaces in YouTube search, in some Google Video results, and increasingly in AI-search citations of video content. Captioning a long-form video is the cheapest way to turn an hour of speech into text the search index can read.

Sources

We drew every figure on this page from a primary source, not a secondary summary:

Drop in a URL above.

You'll have a timestamped transcript in under a minute. From there, conversion to SRT/VTT is mechanical – or see our pages for YouTube videos, podcast episodes, Instagram reels, and TikTok videos if your source needs different framing.