YazSes

YazSes is an open-source desktop dictation and transcription tool for Linux, macOS and Windows. Hold a key, speak, release — the text is typed straight into whatever window has focus: editor, browser, terminal or chat. After installation and the one-time model download, speech recognition runs locally on your CPU via faster-whisper, so audio and transcripts are not sent to a cloud service.

Beyond dictation it transcribes existing audio and video files to text, Markdown, SRT, WebVTT or JSON, and can capture a whole meeting hands-free and produce a speaker-labelled transcript — optionally with minutes (summary, decisions, action items) written by a local language model. Speaker names come from voiceprints you enroll yourself, stored encrypted on your own machine.

It needs no GPU, no account and no subscription, and it is Apache-2.0 licensed. Linux is the primary platform (X11 and Wayland); macOS and Windows are supported but less extensively validated on real hardware. The design and benchmarks are described in an arXiv paper.

Accessibility features include a dysfluency mode that collapses stutters and repeats for stuttered or dysarthric speech, per-user voice calibration, and an alternative activation route using a muscle (EMG) sensor for people who cannot press a key.