Talk2Text

Free & Open Source source available

Dictation that runs on your own machine, translates as it types, and does not ask you to sign in to anything. Built because every good speech-to-text tool wanted an account, a subscription, or an upload of my voice to somebody else's server.

Get it on GitHub Get the weekly update

Video Screen recording: speaking a sentence in Spanish and English text appearing as you go

Why it's interesting

The parts worth showing someone

Talk in one language, get text in another

Translation happens in the same pass as the transcription, so there is no second step and nothing to paste between two tools.

The model runs on your machine

No upload, no account, no per-minute pricing. Your voice never leaves the machine, which also means it keeps working with no internet at all.

It picks the model that fits your hardware

It checks what GPU you have and chooses a model sized for it, so a laptop and a workstation both get the best they can run. Nothing to compare, nothing to configure.

Desktop and phone

The same tool in both places, so notes dictated on a walk land in the same format as notes dictated at a desk.

How I built it

The decisions behind it

The whole design follows from one decision: run the model on the user's machine. That rules out the easy path of calling a hosted API, but it also removes accounts, billing, rate limits, and every privacy question anyone could reasonably ask.

The problem that creates is hardware. A workstation with a large GPU and a laptop with none cannot run the same model, and asking a user to know which one to pick is asking them to care about something they should never have to think about. So it reads the GPU and chooses a model sized to it. That is why there is nothing to configure.

It is not code-signed yet, which is the honest rough edge. Windows shows a publisher warning and macOS Gatekeeper blocks it until you approve it manually. Signing certificates cost money and mean identity paperwork, so for now the trade is: free, open, and slightly awkward to install.

See it

What it looks like

Screenshot The main window mid-dictation, with the language pair visible

About this one

Questions people ask.

Will it run on my machine?

It checks what GPU you have and picks a model sized for it, so a laptop with none and a workstation with a big one both get the best they can run. There is nothing to compare and nothing to configure.

Why does my computer warn me when I open it?

Because it is not code-signed yet. Windows shows a publisher warning and macOS blocks it until you approve it by hand. Signing certificates cost money and require identity paperwork, so for now the trade is free, open, and slightly awkward to install.

Does my voice go anywhere?

No. The model runs on your machine, so there is no upload, no account and no per-minute pricing - and it keeps working with no internet at all.

Is it really free?

Yes, and open source. There is nothing to sign up for and nothing to pay.

Next step

Want to know what lands next?

One email a week: what shipped, what's next, and what's happening.

Get the weekly update See the community

Other things I'm building

See all 6

Builder Console

Dev Tool

All your terminals in one app, easily organized.

built, not released Open →

Cluster Alpaca

Game

A finished mobile game you can install on Android today.

android apk Open →

Math Dungeon

Educational Game

Kids build a dungeon, then play it - and every trap is maths.

built, not released Open →