Get Voice Pro

Get Voice Pro Voice Pro is the speech-to-text that never sends your words anywhere. It types where your cursor is, it understands your accent, it translates as you speak, and it can sit through a two-hour meeting and hand you a transcript afterwards — all on the machine you already own. Your voice never leaves your computer. That is not a feature of the product; it's the whole point of it.

Linux

The Linux build is a single portable file. Download it, make it executable, run it. No installer, no packages, no repository.

Download Standalone Binary

Check one thing before you download: it needs glibc 2.38 or newer, and that's a hard floor — set by how it's built, not something to work around. If your system is Fedora 39 or later, Ubuntu 24.04 or later, Debian 13 or later, Arch, or CachyOS, you're fine. Anything older, and the file won't start at all.

v2.3.3 · ~161 MB · Models download on first run (~1–3 GB)

Windows

Windows: a normal installer. Windows gets a small installer that doesn't require Administrator rights.

Download .exe Installer

Run it, click through, and Voice Pro is on your machine. It runs on Windows 10 or 11, and it has exactly the same features as the Linux build — nothing held back, including translation.

v2.3.3 · ~118 MB · Models download on first run (~1–3 GB)

System requirements: the laptop you already have

Component Minimum Recommended
OS Linux (glibc 2.35+) or Windows 10 Ubuntu 22.04+ / Windows 11
RAM 4 GB 8 GB+
GPU Not required (CPU-only mode) Any Vulkan/CUDA GPU with 2+ GB VRAM
Disk 2 GB (app + Fast Light model) 5 GB (app + all model tiers)
CPU x86_64 with AVX2 4+ cores, 3 GHz+

Quick Start

1

Download and install

Pick your platform above. On Windows, run the installer. On Linux, make the file executable and run it. The first time you do, you'll be asked for an email — nothing else. That gets you the 14-day trial, with no card required. It unlocks everything except personal training and the agent connection.

2

2. Launch and download a model

Once it's running, download a transcription model. That's the one transfer that takes a while — a few minutes, once. Start with the small model and switch to the larger one if your audio is noisy or accented. Translation models are optional; add them whenever you need them.

3

3. Press Ctrl+D and speak

Your words land where your cursor is — an editor, a terminal, a chat box, any field that takes a keyboard. If it hears a word wrong, fix it once and that correction becomes part of your model; the word comes out right next time. No sample sentences to read out, nothing to babysit. And everything you just said went from your mouth to your keyboard — nowhere else.

Hear it when it ships

New releases, real benchmarks and the occasional deep-dive. No spam, unsubscribe in one click.

Everything we build

External:   YouTube · GitHub