open source · MIT · 100% local

Press a hotkey. Speak.
Your words appear.

Local, GPU-accelerated voice dictation. Whisper runs on your hardware — no cloud, no subscription, no audio ever leaving your machine. Built for the platform the paid apps skipped: Linux. Now on Windows too.

0.04 sto transcribe 4 s of speech (RTX 3070)
$0forever — your GPU, your model, MIT license
0 bytesof audio leave your machine
96hardware-free tests run in CI on every commit

Live preview

See what it hears, while you're still talking.

A minimal always-on-top card shows the transcription as it forms — and never steals focus from the window you're dictating into. Release the hotkey and a final, full-context pass is pasted at your cursor.

listening

Faithful replica of the real overlay — matte, monochrome, focus-proof.

Features

Everything a dictation tool should do.
Nothing phoning home.

Instant on your GPU

A daemon keeps Whisper large-v3-turbo warm in VRAM: the text lands the moment you stop speaking. No GPU? Automatic CPU fallback — slower, still private.

Live preview

An on-screen card shows the transcription while you speak, refreshed every second — engineered so no compositor can hand it focus.

Voice-chat aware

Dictating mid-call? Your mic stream to Discord is muted — the call never hears your prompts — and the call's audio is ducked so it stops derailing your sentence.

Any language, your jargon

Dictate in whatever Whisper speaks. Feed your product names into the custom vocabulary and the decoder stops mangling them — it biases, never rewrites.

History & statistics

Pasted into the wrong window? Every dictation is kept locally — recover it with voiceflow last --copy. Words per day, charts, a GitHub-style activity calendar.

Private by design

The model, the audio, the history, the config — all on your disk. The only network call in the codebase is an optional once-a-day update check. That's it.

Why voiceflow

The paid apps are great.
They just aren't yours.

voiceflowtypical paid dictation app
pricefree, MIT~$15/month
audio leaves your machineneveralways
Linux / GNOME / Waylandnativeunsupported
latency after you stop speaking~0.1 s on a GPUnetwork round-trip
works offlineyesno
source codeall of it, on GitHubclosed

Install

One command. Everything included.

The installer sets up the environment, the hotkey, the background service, and downloads the speech model with a visible progress bar — when it says done, dictation works. Updating later? Run the same command again.

Linux Ubuntu · GNOME on Wayland

One step asks for sudo (input tooling + udev rule — the script explains why). Press Super + G, speak, press it again.

curl -fsSL https://raw.githubusercontent.com/AveJaPl/voiceflow/main/install.sh | bash

Prefer to read it first? View install.sh on GitHub.

Windows 10 / 11 beta

Works in PowerShell and cmd, from any directory. Press Ctrl + Shift + Space, speak, press it again.

powershell -NoProfile -ExecutionPolicy Bypass -Command "irm https://raw.githubusercontent.com/AveJaPl/voiceflow/main/windows/install.ps1 | iex"

No terminal? Download the double-clickable installer:

voiceflow-install.bat

Details, requirements and current limitations: README · Windows notes · all releases

Platforms

Every platform. That's the plan.

Linux — GNOME/WaylandThe birthplace. GPU pipeline, overlay, ducking, GTK settings app.live
Windows 10 / 11Same core, Win32 layer: global hotkey, clipboard paste, audio ducking.beta
KDE · Hyprland · SwayEasiest port — these compositors have the protocols GNOME lacks.planned
macOSAVFoundation recording, CGEvent paste, Apple Silicon inference.planned
AndroidA system keyboard (IME) with on-device Whisper — dictate in every app.in progress
iOSKeyboard extension, sibling of the Android app.planned

voiceflow stays free and open source — on every platform we can reach.

Good dictation shouldn't cost $15 a month or require shipping your voice to someone else's server. If you want it on your platform sooner, the issues above have implementation notes waiting — grab one, comment, and go. Contributions welcome.