More than a wall of text
Every transcript comes back as separate paragraphs — one per audio segment Whisper detects — with the segment's start time stamped on the left in monospace. You can scrub to a moment, copy a quote, or download the whole thing as a plain .txt without losing the timing markers. If you want those timings as subtitles rather than prose, the same transcript exports as .srt and .vtt caption files. OpenAI's Whisper paper (Radford et al., 2022) reports the model was trained on roughly 680,000 hours of multilingual speech across 99 languages — that is the same family running here, in its large-v3-turbo form, so the free tier stays both accurate and free.
How it works (1 minute, 3 steps)
- Paste the YouTube URL. The same link YouTube's Share menu gives you. Long-form videos, Shorts, and unlisted-but-publicly-shareable links all work. Private and members-only videos cannot be reached anonymously — that is a YouTube rule, not ours.
- We pull the audio. shrybe extracts just the audio track from the source URL. That usually takes 5 to 15 seconds. We never download the video file itself; only the audio Whisper needs.
- Whisper writes it down. The audio runs through OpenAI Whisper on our backend. For a 10-minute talk you typically see the first paragraph in your browser inside 30 seconds. The result comes back over a live connection that stays open, so a slow network cannot drop it on you.
Why not just use Tactiq, Otter, or a Chrome extension?
Tactiq, the most common free-tier alternative, advertises 700,000+ users and runs as a Chrome extension; you install it, sign in with Google, and click an icon on the YouTube page. Otter and Descript are full SaaS products with monthly minute caps and an account requirement. Every one of them is a sound choice for somebody — we picked different trade-offs:
- Nothing to install. No extension, no desktop app, no browser permissions to grant.
- No account. Free tier is rate-limited by IP at 15 requests per day. We never see your email.
- Broader source list. The same form transcribes podcasts, TikTok, Twitter, and a long tail of public-URL audio sources — not only YouTube and meeting platforms.
- Honest about the model. We tell you which Whisper checkpoint produced the text on each result. Most competitors keep that opaque.
If you mostly transcribe meetings recorded inside Google Meet or Zoom, Tactiq or Otter will fit you better — that is their core use case and they are good at it. shrybe is built for the "paste a public URL, get text" workflow.
Who actually transcribes YouTube
The AI transcription market grew from $4.5 billion in 2024 to a projected $19.2 billion by 2034 — roughly 25% compound annual growth, faster than most adjacent SaaS categories. Inside that demand, four audiences dominate what we see in our logs:
- Podcasters and creators pulling rough drafts to caption clips, build show notes, or repurpose a 40-minute episode into a written essay. Episodes that live somewhere other than YouTube go through podcast transcription instead, short vertical cuts through Instagram reels and TikTok videos, and anything not in English can come back translated. If you are holding a phone rather than sitting at a desk, so does transcribing on iPhone.
- Journalists and researchers who need a quotable transcript of a public talk or interview without manually typing for an hour. Court transcripts and YouTube investigative reporting are common.
- Accessibility users — Deaf and hard-of-hearing viewers, and people who simply read faster than they listen. Captions on YouTube are uneven; an honest transcript fills the gap.
- Developers and ML engineers evaluating whether Whisper output is good enough for their own pipeline before building anything themselves.
Audience two and four overlap with what Mayer-Stoye and colleagues (2025) describe as "complex products serving technically savvy audiences," a segment that captures roughly 17× the baseline share of AI-search citation traffic compared with generic consumer tools. If your reason to transcribe a video is "I want to verify what someone actually said," shrybe is squarely aimed at you. And when you want the gist rather than the record, a transcript is the honest first half of summarising a video.
Languages and the Slavic problem
Whisper covers 100 languages on the large-v3 line running here — the 99 of the 2022 paper plus a Cantonese token added in large-v3. The human comparison people like to quote is narrower than it sounds: the paper measured English only, against professional transcription services on 25 recordings, and found Whisper landed within a fraction of a percentage point of pure-human transcription while a computer-assisted service beat it by 1.15 points (Radford et al., 2022, Figure 7). No equivalent human comparison was published for the other 99. The honest weakness is Ukrainian Surzhik, Belarusian, and short clips in any low-resource Slavic language: Whisper's own automatic language detector runs on a few seconds of audio and confidently picks Russian for clips that are actually Ukrainian, producing what looks like real Russian words mapped to wrong Ukrainian sounds.
We work around that by pre-detecting language from the source title, description, and
uploader name before Whisper ever runs. A Cyrillic YouTube title containing
і, ї, є, or ґ is a near-perfect
Ukrainian signal. If our heuristic finds one, the language picker auto-selects
"Ukrainian" with a "detected from source" hint, and Whisper runs in
Ukrainian-only mode. You can override the pick by clicking the dropdown — your manual
choice is always preserved.
What we do with your data
The audio is fetched from YouTube to our backend, runs through Whisper, and the resulting text is returned to your browser over a live connection. We do not store the audio long-term. Cached files older than 24 hours are deleted in a nightly job. We do not sell, share, or train any model on what passes through the free tier. There is no account, so there is nothing to link a transcript back to a person.
The free tier rate-limits by IP at 15 transcripts per 24-hour window. We log the IP address and the source URL purely to enforce the rate limit and to debug failures; both rotate out within 30 days.
Frequently asked
Do I need a Chrome extension or an account?
No. shrybe runs in any browser tab on any device. There is no extension to install, no signup, no email needed. The free tier allows 15 transcripts per IP address per day, with no rate-limit prompt before you use them.
What kinds of YouTube videos can I transcribe?
Any public YouTube URL, including Shorts, livestream replays, and unlisted-but-publicly-shareable links. We cannot reach private or members-only videos because YouTube blocks anonymous access to those. Age-restricted videos sometimes work, sometimes not, depending on YouTube’s current bot policy.
What is the maximum video length?
The free tier handles videos up to about 75 minutes — long enough for most talks, interviews and full podcast episodes. Longer videos require more compute and we cap them to keep the free service free. For anything past that, split it before pasting, or wait for the paid tier we are still designing.
Which transcription model are you using?
OpenAI Whisper. The free tier runs the large-v3-turbo checkpoint — a distilled large-v3 that keeps most of its accuracy at a fraction of the compute. The much smaller "small" checkpoint is used in exactly one place: the iOS Shortcut endpoint, where a hard timeout forces a faster model. Full large-v3, still the leader on Slavic and low-resource languages, is on the roadmap as a paid tier.
How does the language picker work?
Three layers. First, we default the picker to your browser locale. Second, when you paste a URL we silently fetch the video's title, description, and uploader name, and run a script-detection heuristic against that text — a Ukrainian title in Cyrillic with letters like “і”, “ї”, “є”, “ґ” is the strongest possible signal that the audio is Ukrainian, far better than running Whisper detection on a noisy 30-second clip. Third, if you pick something manually, we never override your choice.
Are timestamps included?
Yes. Every transcript renders as one paragraph per audio segment with the start time in monospace on the left. Copy and Download keep the same paragraph structure so you can paste straight into a doc or feed it to another tool.
Where does my data go?
The audio is fetched from YouTube to our backend, run through Whisper, and the result is streamed back to your browser live. We delete the cached audio and the transcript text within 24 hours. We do not share or sell anything we transcribe.
Why does the result sometimes look like garbled Russian on Ukrainian clips?
Whisper's automatic language detection runs on a few seconds of audio and frequently misreads Ukrainian Surzhik as Russian, which produces real-looking Russian words that map to wrong Ukrainian sounds. Our metadata-based pre-detection (see above) sidesteps that for most URLs. If it still happens, manually pick “Ukrainian — Українська” in the language picker and rerun — an explicit language argument skips auto-detection entirely and returns a correct transcript.
Ready when you are.
Drop a YouTube link into the box at the top of this page, or go back to the homepage if you also have podcasts or audio files to transcribe.