How to use
🎙 For live streams (live captions in OBS)
- Choose “For live streams” at the top and set the caption look (template, font, colors).
- In OBS, open “Tools” → “WebSocket Server Settings”, enable the server, enter its password here and press “Connect to OBS”. The password is remembered on this device only, so it connects automatically next time (it is never sent anywhere; press “Forget” to remove it).
- Paste the address shown into an OBS “Browser” source (width 1920, height 1080).
- Press “Start live captions” and talk. Captions appear in OBS a few seconds later (the measured delay is shown). Need frames for your scene? Make them in the overlay maker.
🎬 Caption a video
- Choose “Caption a video” at the top and add a recording or video file.
- Pick the spoken language and press “Transcribe”. The AI downloads once the first time.
- Click any line to play from that moment, and fix the text while you listen.
- Set the caption look and shape (landscape, vertical, square), then save SRT or burn the captions into the video. The full version adds VTT and ASS.
Why “no upload” matters
Most transcription services send your recording to their servers. This tool runs the AI (Whisper) inside your browser, so your audio and text never leave your device. The only network use is downloading the AI files once (from Hugging Face and jsDelivr) and checking your license key or trial.
How “no upload” works
Everything is processed on your device. Zero network requests are made while transcribing. A “network log” appears after each run so you can check it every time.
Every server this page talks to
| Server | When | What is sent |
|---|---|---|
| models.chocotto-games.com (our AI file host) | First time only (AI files) | Normal access data only (never your audio or text) |
| cdn.jsdelivr.net | First time only (AI runtime) | Same as above |
| chocotto-games.com | Opening the page, license and trial checks | Your license key and a random device ID only |
| fonts.googleapis.com etc. | Opening the page | Loading fonts |
| Google ads (googlesyndication.com etc.) | While the page is open | Showing ads; never your audio or text |
What the license check sends
When you enter a key, and every 30 days after, we send only the key and a random device ID created in this browser (to count devices). Our server checks the key with Gumroad and returns the full-version code. Your audio, text and file names are never sent.
Check it yourself
- Open this page in Chrome or Edge on a computer, press F12 (⌥⌘I on Mac) and open the “Network” tab.
- After the AI has loaded once, clear the list (the circle-with-a-line button) and press “Transcribe”.
- You will see no requests carrying your audio while it transcribes (ad requests may appear).
- For extra certainty, turn off Wi-Fi after the AI has loaded, then press “Transcribe”. It works offline.
Plans and features
| Free | Trial (once) | Creator | |
|---|---|---|---|
| Price (one-time) | $0 | $0 | $39 |
| Transcription | Up to 3 min, standard AI | One file up to 60 min, high-accuracy AI | Unlimited, high-accuracy AI (Whisper large-v3-turbo) |
| Live captions (streaming, OBS) | Up to 10 min, with logo (Auto, Fastest or Balanced AI) | No time limit, no logo, plus the “Most accurate” AI | |
| Overlay maker | Templates, themes, readability check, PNG / WebM / HTML for OBS (watermarked) | Everything on the left (no watermark) | |
| Video subtitles | Subtitle designs, vertical and square, burn-in, SRT (small watermark in a corner) | Everything on the left (no watermark) + word-timed splitting, VTT, ASS, English subtitles | |
| Handy tools | Search, edit while listening, auto-save | Everything on the left + word list, speaker separation, silence skipping, recording, highlighted low-confidence lines | |
| Ads | Yes | None (devices with a key never load the ad code) | |
Names and technical terms may come out wrong. Add them to the word list and they will come out right.
Founder price for the first 50 buyers: $29. Register up to 3 of your own devices, one active at a time.
The free version lets you try every subtitle design, live captions and the overlay maker (watermark or logo). For a long recording, “Try the high-accuracy AI once” transcribes one file up to 60 minutes.
More: For creators (live captions, subtitles, vertical exports).
Updates
Bug fixes and small improvements are always free. Major new features (for example live subtitles, translation into Asian languages, or multi-clip exports) may be sold separately, with a discount for existing buyers. If a browser change breaks a feature, we will work to fix it.
FAQ
Is my audio really not uploaded?
Yes. Transcription happens only in your browser. After the AI has loaded, it even works offline.
How fast is it?
On Chrome or Edge on a computer (WebGPU) it is faster than real time. Without WebGPU, or on phones, it is slower and best for short clips.
How big is the AI download?
About 80 MB for the standard AI and about 600 MB for high accuracy. It downloads once and is kept by your browser.
Names and terms come out wrong
Add “wrong -> right” pairs to the word list in the full version to fix them all at once.
Can it transcribe other languages?
Yes. Choose “Other” to pick Korean, Chinese (Simplified or Traditional), Spanish and more, or let the AI detect the language automatically. Translation is available for Japanese audio to English.
If I make English subtitles, are the Japanese ones kept?
No. Translating replaces the subtitles with English only. To keep both, make the Japanese subtitles first and export them (SRT, etc.), then run it again in English. You get two separate files.
How many devices can I use?
Every plan: register up to 3 of your own devices, one active at a time. Entering the key on another device switches the full version there, and the previous device goes back to the free version (press “Use on this device” to switch back any time). For company-wide use, please contact us.
Is it a subscription?
No. It is a one-time purchase. Bug fixes and small improvements are always free.
Can I get a refund?
Refunds within 7 days of purchase if you have not activated the key yet. After activation, no refunds as a rule (except when it does not work because of a problem on our side that we cannot fix; refunded keys stop working). Before buying, please check with the free version (up to 3 min) and the “Will it work on this device?” check.
AI models and licenses
Whisper (OpenAI, MIT License) / opus-mt-ja-en (Helsinki-NLP, Apache License 2.0) / Silero VAD (MIT License) / opencc-js (MIT License) / WeSpeaker (CC BY 4.0) / Transformers.js and ONNX Runtime Web (Apache License 2.0, MIT License).