Why Voco exists
A new machine revealed a missing gesture.
I switched to Linux a few weeks ago after becoming disappointed with Apple and its MacBook Pro. These used to be the creme de la creme when it came to high end laptops. I do not believe that still holds true anymore. These machines are very heavy, no longer as aesthetically impressive as they once were, and still are very expensive. I decided to try out a Lenovo ThinkPad because of my keen interest in experimenting with Linux. I love it, and with the current advancements in agents operating in CLIs, I now have a machine that is very powerful at a third of the price with assistants helping learn how to use it and build things for it.
The desktop had intelligence. It lacked a native voice gesture.
With this new change come new challenges. Not all the applications I was used to, or potential new ones, are immediately available. That can also be seen as a great opportunity to learn and build.
The idea
What if speaking felt native?
This is where Voco, my latest mini project, enters the conversation. I loved seeing how people are now dictating to their CLIs instead of typing, not just to CLIs, but in fact to a lot of applications. I really liked the idea and wanted to give it a try. To my surprise, Whisper Flow is not yet available on Linux. But what if I could build it myself? That could save me a few pounds at the expense of some precious tokens.
- Voicemicrophone
- Audio samples16 kHz mono
- Local transcriptwhisper.cpp
- Focused fieldIBus v3
Under the surface
Default dictation stays inside the laptop.
Voco is a free, open source desktop dictation tool for Linux. You press Alt+D, speak, and your words appear wherever your cursor is. Everything runs locally using whisper.cpp.
- CaptureWeb Audio · 16 kHz mono
- Native engineTauri 2 · Rust · whisper.cpp
- Desktop deliveryIBus v3 · Wayland / X11
The cursor contract
Fast while speaking. Exact when committed.
A revisable preview follows the cursor. Voco makes each chunk permanent only after the focused field passes an exact-prefix check; otherwise, the transcript stays in Voco.
your words can still change here
30 s windows · 1 s overlap · 29 s stride · fail closed