About
Hold a key, speak, release, and your words are typed into whatever app you are using. Speech recognition (faster-whisper, on your CPU) runs on your own computer, so your voice never leaves it. No account, no subscription, no API key. Free and open source under Apache 2.0.
Voice commands such as "scratch that" and "new line". Offline transcription of audio and video files with speaker labels. Whole-meeting capture with a speaker-labelled transcript. Works under Wayland on Linux.
Measured, not claimed: 4.07% word error rate on LibriSpeech test-clean (base.en, int8, laptop CPU). Clean read speech flatters every system, so real dictation will be worse; the method is published.
Linux is the most tested platform. macOS and Windows builds exist but have had less hardware testing.
Install: snap install yazses, or pipx install yazses. Source: https://github.com/MSKazemi/yazses