OpenDictate v0.1.7
OpenDictate — Local-First Voice Dictation
Highlights
-
Dual-capable installers (Linux & Windows): One download now carries a CUDA-capable speech runtime alongside the classic CPU path. Have an NVIDIA GPU? Flip on Settings
$\rightarrow$ GPU acceleration and the Models page shows aGPU ✓chip; every other machine behaves exactly as before with silent CPU fallback. -
macOS stability wave (community PR #1): No more UI freezes on start/stop; fixed a crash that killed the app right after every successful paste-insert; paste/undo now
⌘V/⌘Z; native notifications without a"Where is use_default?"picker; onboarding rebuilt withEsc/$\times$ dismissal; double-tap modifier shortcuts (Fn Fn,⌘ ⌘, etc.).
Changes since v0.1.6
- GPU execution-provider plumbing across the STT stack with automatic CPU fallback (live-tested on non-NVIDIA hardware) and truthful provider reporting.
-
Settings
$\rightarrow$ GPU acceleration: Off / Auto / CUDA / CoreML (experimental). - Default macOS shortcut is now
⌘ ⇧ Space; dev hot-reload works again. - README lists all models with accurate sizes;
docs/GPU.mddocuments the verification matrix.
📦 Installers
| Platform | File | Notes |
|---|---|---|
| Linux | .deb, .AppImage |
CUDA runtime bundled (~+150 MB); CPU fallback everywhere else |
| Windows | .exe, .msi |
CUDA DLLs installed alongside (~+300 MB) |
| macOS | universal .dmg |
Apple Silicon + Intel; CoreML pending validation |
Note: Models download inside the app on first run; everything stays local. MIT licensed.