-
Notifications
You must be signed in to change notification settings - Fork 4
FAQ
Yes, it's fully offline, and no. Transcription runs locally with faster-whisper; there are no network calls in the dictation path, no API key, and no account. You can confirm with yazses doctor (it reports "fully offline") and by watching your network — there's nothing to see.
- vs. cloud dictation (Wispr Flow, Google/Apple/MS): YazSes runs entirely on-device — nothing is uploaded. That's the whole point.
- vs. Talon Voice: Talon is a powerful scripting/accessibility platform with a learning curve. YazSes aims at plug-and-play dictation + simple voice commands. Different goals; both valid.
- vs. built-in OS dictation: YazSes is cross-platform, open source, private by construction, scriptable, and works into terminals and editors.
It depends on the Whisper model you choose (smaller = faster, larger = more accurate) and your CPU. The defaults target a good balance on a typical laptop CPU with int8 quantization — no GPU required. If dictation feels off, yazses tune and yazses mic-level help you calibrate.
Almost always one of:
-
Mic level below the VAD threshold → run
yazses mic-level --setto recalibrate. -
The OS default mic switched (e.g. a USB-C monitor stole capture) → the mic-change guard auto-heals, or pin your mic with
yazses audio use <name>. -
First words clipped → there's a pre-speech padding setting; see
docs/.
yazses doctor --mic diagnoses most of these.
The defaults are tuned for English (.en Whisper models), but faster-whisper ships multilingual models you can select via config. Better multilingual support and code-switching are active roadmap items — this is a great area to contribute. See Ideas.
Dictation and injection work across X11 and Wayland (with the right injector — yazses doctor checks readiness). Some features that need to focus another window (like look-to-pane targeting) are X11-only today, because Wayland restricts cross-window control. Improving the Wayland story is on the roadmap.
The architecture is cross-platform (shared Protocol interfaces with per-OS backends) and CI runs on all three. Linux is the most battle-tested day to day; helping harden macOS/Windows is high-value and very welcome — see Contributing.
YazSes has a large catalogue of advanced, opt-in capabilities (voice commands, meeting capture, offline file transcription, accessibility modes, and experimental perception features). They stay dormant until you run yazses features enable <name>, so a fresh install is simple and nothing changes your setup unless you ask. Browse them with yazses features.
Start at Contributing — Start Here. Pick a good first issue, or open an Idea. Small PRs and docs fixes are genuinely welcome, and we review fast.
- 🐞 Bug → open an issue with your OS,
yazses --version, andyazses doctoroutput. - ❓ Question / how-to → Q&A discussions.
- 🔐 Security → see
SECURITY.md; please don't file security issues publicly.
YazSes — offline, on-device voice dictation · Apache-2.0 · Docs · Discussions · Good first issues
Contribute
Use it