Launching today
TrainScription
Local AI transcription — ANY MEETING APP. No bot. No cloud.
10 followers
Local AI transcription — ANY MEETING APP. No bot. No cloud.
10 followers
Whisper runs on your device. No bot joins your call. No audio hits a server or saved — ever. Works with Teams, Zoom, Discord, and any app making sound on your machine (Full Desktop mode, Pro). Not just browser tabs. The Phonetic Brain permanently fixes the words AI always gets wrong — your names, your jargon, your clients. Train it once; it corrects automatically on every future transcript. Free tier included. Pro is $9.99 once. No subscription. All future updates included.
Hey PH 👋
🚨 v7.5 ships Full Desktop mode. This is the big one.
Browser-tab transcription was always just the starting point. Full Desktop mode captures ALL sound from your machine — MS Teams desktop, Zoom, Discord app, Slack, anything making a sound. Not just what's in a Chrome tab. Select your screen in the OS picker, enable system audio sharing, and TrainScription captures the full dual-channel stream the same way it always has.
✨ The Phonetic Brain is the feature I'm most proud of.
Whisper is great at general language and terrible at your language — your client surnames, your case names, your product names, your campaign terms. Generic AI transcription gets these wrong every single time, in predictable patterns.
The Brain fixes this permanently. Highlight a misfire in the transcript → assign the correct spelling → done. That word is now corrected automatically on every future transcript, before you see it. The more sessions you run, the sharper it gets. It's yours — portable as a `.train.json` file you can carry between installs.
Pricing: Free tier is genuinely useful (15-min sessions, 3 Brain slots, real transcription). Pro is $4.99 for a limited time — one payment, no subscription, works across devices via email login, all future updates included.
---
Why I built this: I kept running into three problems with every transcription tool I tried:
A bot joins your call — visible to everyone in the meeting, awkward in sensitive conversations, and against some platform TOS.
Your audio goes to someone's server — which means someone else is processing what you said, indefinitely.
My fantasy roleplaying game 😉 names, locations, etc. were completely mangled with generic transcriptions — this was the catalyst of my realization that this issue extends to other jargon-heavy domains.
TrainScription solves all three. OpenAI Whisper runs locally via WebAssembly. Your audio never leaves or is saved on your machine. No bot. No cloud round-trip. Works fully offline after the model downloads once.
Happy to answer anything about the architecture, the Brain system, the privacy model, or why I chose to build this in a Chrome Extension instead of a desktop app. 🎙️