TLDR:
- Best macOS, Linux, and Windows app: Handy.computer
- Alternative for macOS and Linux: Voxtype
- Best model for multiple languages like English and Portuguese: parakeet-tdt-0.6b-v3-int8
I’ve been looking for a good speech to text solution that was offline and open source, and could work across multiple operating systems. Watching one of The Download videos from GitHub, I learned about Handy.computer. It is a great, easy-to-use app that you can set up quickly, and it has some great settings, like configurable shortcuts (with a push-to-talk option), removal of filler words, and custom words (like Kossmann). It even has post-processing with another LLM if you want a more polished result. I’ve been using it for several months on macOS and it has been great!
The second challenge was finding a good model that was fast, accurate, and supported English and Brazilian Portuguese. The Handy.computer documentation page about models is a great reference, and after doing some testing, I found the best results with the model parakeet-tdt-0.6b-v3-int8 from NVIDIA. Even though it only officially supports Portuguese from Portugal, it does a good job with Brazilian Portuguese too.
I had some trouble getting Handy to be triggered by the hotkey on Ubuntu. Maybe I was missing a dependency or something like that, so I was only using it on my work computer (macOS). Since I switched to Omarchy on my personal computer, which comes with Voxtype pre-installed, and it’s working great! I miss the easy-to-set-up interface from Handy, but it has been a good experience trying an app that has a different way of working. I still need to set up custom words (it’s not as straightforward as it is on Handy), but I’m really happy with how well it works.
I’m still a person who thinks better when I write, but for short replies or for giving feedback to an AI agent, speaking is much faster than typing, and using a speech to text app for these specific use cases has been a great experience. And if you’re wondering, this post was completely typed.

Leave a Reply