Due to an accident, I’m temporarily down to one usable arm which makes my typing rather slow. Wondering if someone has some form of speech-to-text for text entry done in NixOS. I’ve found stt but it seems rather bare and just a component of a workable solution. I’d possibly imagine like a virtual keyboard or hotkey, then I say my stuff, and then the recognized text gets typed like a virtual keyboard. Also open for other suggestions. I can do config changes but anything that still requires a lot of experimentation is out of the picture for practicability reasons.
I tried packaging three different tools but I was not impressed by either of them:
https://github.com/NixOS/nixpkgs/pull/185255
https://github.com/NixOS/nixpkgs/pull/173847
https://github.com/NixOS/nixpkgs/pull/185148
In general there is talon, which I managed to package in an earlier version, though not the current one. Though the old version isn’t available anymore.
It’s much more than just TTS, but a full command engine and can therefore be used for programming as well.
If someone manages to package it stably, I’d like to hear more about it!
I found Talon as well. Managed to get a current version to run using steam-run. Unfortunately, there’s still a lot of wrong detections and sometimes trying to correct using my voice makes everything worse. However using my remaining hand to control cursor and special keys is quite a good combination and also quite fast. I will play around with this some more and see if I am actually faster than typing one handed. Will definitely try to create a Nix package for this once I’m back to two typing hands.
What is the status of this. I would be interested in using nerd-dictation, but haven’t got the first clue how to get it installed.
Someone needs to adopt the package and create a PR against Nixpkgs, addressing the review comments.
There’s also wyoming-faster-whisper which is popular in the Home Assistant community, runs locally, works really well, and already packaged for NixOS.
I’m able to run it, but haven’t figured out how to pipe output to file.
Hey,
Inspired by people vibecoding on youtube using Wispr Flow, I wanted something similar on my amazing nixos setup. I made WhispAway, a small rust wrapper around whisper.cpp and faster-whisper for push-to-talk real-time speech-to-text transcriptions.
If you have a machine with a GPU it should be pretty performant. Tested on two machines, one with an Nvidia chip and one with a Meteor Lake inegrated Intel Arc GPU. Currently packaged with wtype for text output, which I guess makes it usable only on wayland currently.
This is great. I just stumbled onto your app and it is working fine.
Can you add an option to also copy the text into the clipboard?
Happy to hear someone more than me gains benefit from it. I’ve now added the clipboard option.
Hey, fantastic!! I’ll check it out.
I also made this speech to text CLI tool written in Rust: Meet Jambi - a blazing-fast voice transcription application built with Rust