Skip to content

v0.10.1 — On-device voice dictation + dictation pill

Latest
Latest

Choose a tag to compare

View all tags
@github-actions github-actions released this 28 Jun 20:39
v0.10.1
2f4f997

🛠️ This is the working build of the 0.10 voice release. v0.10.0 failed to start once installed (a packaging bug left two main-process modules out of the bundle); v0.10.1 fixes that. If you grabbed 0.10.0, install this instead.

🎙️ DockShift can now hear you — fully on-device

This release adds local voice dictation as a headline feature. Speech-to-text now runs entirely on your machine — your audio never leaves the device, there's no cloud round-trip, and no API key is required.

✨ What's new

  • On-device voice dictation. Powered by sherpa-onnx running NVIDIA's Parakeet TDT 0.6B v3 model — 25 European languages, auto-detected, running on your CPU. This is now the default speech engine for every install.

  • Dictation pill — press Ctrl+Shift+Space anywhere. Hit the global hotkey, speak, and your words are transcribed on-device and auto-pasted straight into whatever app is focused — your editor, chat, browser, anything. A floating, draggable, non-activating pill shows your live mic level and status, then dismisses itself. The hotkey is rebindable, and you can switch the output between auto-paste and copy-to-clipboard in Settings → Shortcuts.

  • Pick your engine on first run. The welcome flow now has a voice-engine step with at-a-glance accuracy/speed bars and a multilingual badge; your chosen model downloads quietly in the background.

  • Verified model downloads. Speech models are SHA-256 verified before first use, extracted atomically, and the engine warms on demand and idles back down when unused.

🔁 Changed

  • The default speech engine is now on-device Parakeet (previously the in-browser Vosk small model, which was too inaccurate to be useful).

🗑️ Removed

  • The legacy Vosk engine is fully retired — provider, in-browser WASM, the vosk-model:// protocol, the vosk-browser dependency, and its CSP allowances.

🐛 Fixed

  • Packaged app failing to start (ERR_MODULE_NOT_FOUND) — electron-builder's file allowlist is now a glob, so every main-process module ships in the bundle. Verified by launching the packaged executable.
  • Upgrading no longer leaves you stuck on "Loading speech model…". Anyone carrying a saved legacy engine setting is automatically migrated to on-device Parakeet on next launch, and the old Vosk model files are cleaned up to reclaim disk.

📦 Install

  • New users / clean install: download DockShift-Setup-0.10.1.exe (one-click installer).
  • Portable / no install: download DockShift-0.10.1.exe and run it directly.
  • Existing users: auto-update will pick this up; the first time you use voice, the ~600 MB Parakeet model downloads once in the background.

ℹ️ Windows 10/11 only. The installer is not yet code-signed, so SmartScreen may warn on first run — choose More info → Run anyway.

Full Changelog: v0.9.4...v0.10.1

Assets 6
Loading