No bot. No cloud.
Any app. Just the words.
Real-time transcription that runs entirely on your device — browser meetings or desktop apps. Speaker-labeled, timestamped, and improved over time by a vocabulary Brain you train yourself.
Free tier included · Pro unlocks unlimited sessions, unlimited Brain slots & more at $9.99 — once, forever
Mint-highlighted words were corrected by the Phonetic Brain — Whisper heard them wrong; the Brain fixed them. Highlights appear live in the Studio as each correction fires.
By the numbers
How it works
TrainScription captures audio from your machine's audio pipeline — not from the meeting's participant list. Browser Tab mode works with Google Meet, Teams (web), Zoom (web), Webex, Discord, and any other browser-based call. Full Desktop mode captures all system audio — Teams desktop, Zoom, Discord app, or anything making sound on your machine. Free tier sessions are subject to the 15-minute cap. No third-party participant joins the call in either mode.
Select Browser Tab to capture a meeting in Chrome, or Full Desktop to capture any app making sound on your machine. The extension captures two audio streams: your microphone and the meeting audio. Both feed into a local Web Audio graph. A silence gate drops empty chunks before Whisper ever sees them. The AI runs entirely in your browser via WebAssembly — no network requests during transcription, no server, no API key.
Every 5 seconds, a new chunk of speech is transcribed and appended — speaker-labeled with your configured channel names and a precise timestamp. Open the full Studio view to see the live transcript, or just leave the popup open in the background. Close the popup at any time — transcription continues.
See a word the AI got wrong? Highlight it in the transcript. A popover appears instantly — assign it to the correct spelling in one click. That phonetic mapping is now permanent. Every future transcript corrects that word automatically, silently, before you ever see it. The transcript stays visible after you end a session — no need to dig through Recovery. Finish your corrections, then clear when you're done.
Long sessions are automatically segmented into sealed chunks. All segments from a single Begin→End session are grouped together in Recovery. Export any segment or the full session as a clean .txt or .srt subtitle file. Review any past segment in the live transcript view to catch and correct misfires after the fact. Rename sessions with custom labels anytime. When you're ready, you bring the AI you already trust — one click from Recovery sends the transcript on your terms.
Audio & Video File Import (Pro) runs an existing audio or video file you already have — a meeting export, a webinar, anything — through the exact same on-device Whisper pipeline. Your file is read locally in your browser and processed entirely on your device, never uploaded. The Phonetic Brain corrects it automatically, the same as a live session. And whenever you want everything off the machine at once — a new computer, a reinstall, just your own copy — Bulk Export, Backup & Restore packages every session and your full Brain into one file, and brings it all back with one click.
Built for
The differentiator
The problem every AI transcript has
AI transcription tools consistently fail on specialized vocabulary — in predictable, repeatable patterns. The Phonetic Brain is the permanent fix. Train it once on a word; it corrects that word on every future transcript automatically, before you ever see the output.
The Phonetic Brain
Whisper — the AI model at the core of TrainScription — is remarkable with general language and terrible with your language. Proper names, product names, case names, medication names, character names, niche jargon: all mangled, all the time, in predictable patterns.
The Phonetic Brain is a local correction layer you build yourself. Train it once on a word; it fixes that word on every future transcript, automatically, before you ever see the output. It compounds — the more you use it, the more accurate your transcripts become.
.train.json, and import it into TrainScription Pro. Your Brain knows the word before the first transcript line appears.Train before the meeting. Refine during the meeting. Keep improving after the meeting.
Everything included
Pricing
No credit card. No account required.
- 15-minute session cap
- 3 Recovery sessions stored
- 3 Phonetic Brain slots
- Real-time dual-channel transcript (Browser Tab + Full Desktop)
- Pause controls
- Export to .txt and .srt after each session
- Session naming in Recovery
- Silence Auto-Stop (5 & 10 min)
- AI Summary Handoff
- Bulk Export — all sessions as one .zip
One payment. No subscription. Works across devices.
- Unlimited session length
- Unlimited Recovery sessions
- Unlimited Brain slots
- Full Desktop mode — unlimited, no session cap
- One-click assembled session export
- Import full Brain vocabulary files
- Silence Auto-Stop — all options incl. Indefinite
- AI Summary Handoff — full multi-segment session
- Audio & Video File Import
- Backup Everything & Restore from Backup
- Cross-device via email login
- Everything in Free, forever
How it compares
| TrainScription | Otter.ai | Fireflies | Rev | |
|---|---|---|---|---|
| No bot joins the call | ✓ Never | ✗ | ✗ | ✗ |
| Audio stays on your device | ✓ Always | ✗ | ✗ | ✗ |
| Works fully offline | ✓ | ✗ | ✗ | ✗ |
| Fixes jargon / names | ✓ Phonetic Brain | Paid only | ✗ | ✗ |
| Pre-session vocabulary training | ✓ BrainTrainer | ✗ | ✗ | ✗ |
| Corrections visible in transcript | ✓ Live highlights | ✗ | ✗ | ✗ |
| Works with desktop apps (Teams, Zoom, Discord) | ✓ Full Desktop mode | ✗ | ✗ | ✗ |
| Import an existing recording | ✓ Same on-device pipeline | Cloud upload required | Cloud upload required | Cloud upload required |
| Back up & restore your full history | ✓ One file, any machine | ✗ | ✗ | ✗ |
| Price | Free / $9.99 once | $16.99/mo | $10/mo | $9.99/mo |
| Works in any browser tab | ✓ | Partial | Partial | ✗ |
Why it exists
Most transcription tools route sensitive conversations through external cloud infrastructure, inject third-party participants into your calls, and charge you monthly for the privilege. The costs compound — financially, architecturally, and in trust.
TrainScription was built around a different idea: transcription should happen locally, under your control, on your own machine — regardless of how long the session runs or how sensitive the content is. The constraint of doing it entirely on-device isn't a limitation. It's the whole point.
Under the hood
TrainScription is a Manifest V3 Chrome Extension running OpenAI Whisper via WebAssembly locally in your browser. The AI model is downloaded once from the model provider and cached — transcription itself occurs entirely on-device after that initial download. TrainScription does not operate server-side infrastructure and audio is processed locally on your device and is not uploaded for cloud transcription.
Audio is processed in 5-second volatile chunks that exist only in RAM. They are never written to disk. Audio is not uploaded for cloud processing. The extension's data policy is architecture, not a promise — the extension contains no code to route your audio externally.
The only external connections in this extension: a one-time AI model download from the model provider, and ExtPay license verification for Pro users. Both are unrelated to your content — your audio, transcripts, and Brain vocabulary are never involved in either connection. TrainScription does not execute remote code or load external scripts that alter extension functionality after installation.
The Phonetic Brain correction function runs synchronously on every transcript line before it's written to storage. Your vocabulary corrections apply before you see the output.
Audio & Video File Import runs Whisper inference inside a dedicated background Worker, separate from the page you're looking at — so processing a long file never freezes or slows down the browser tab. Bulk Export, Backup, and Restore are pure local file operations: reading what's already stored on your device, packaging it, or unpacking it back — no inference, no model, no network connection of any kind, in either direction.
Privacy & data
Why TrainScription requests permissions
TrainScription does not use permissions for advertising, tracking, behavioral profiling, or any purpose beyond the transcription workflow described above.
Say Something
Using the beta? Got a use case I haven't thought of? Found something broken? Leave a note — Terrance reads every one.
