Skip to content

Releases: vlr-code/dictly

Dictly 1.2

Choose a tag to compare

@vlr-code vlr-code released this 03 Jul 13:07

Two changes, both aimed at transcription quality out of the box.

Bundled model upgraded: Whisper Large v3 Turbo (compressed)
The app now ships with large-v3-v20240930 (547 MB, quantized) instead of base — near-Large-v3 accuracy for all languages with ~0.15 RTF on Apple Silicon. This is the model the app used to recommend as a separate download; now it works offline from the first launch. If you never picked a model manually, Dictly switches to it automatically. (The download is bigger for the same reason — ~500 MB.)

New: “Keep microphone warm for instant start” (on by default)
macOS powers the microphone hardware down when idle — the next capture start can take a couple of seconds, which is why the first word of a dictation sometimes went missing. Dictly now keeps the mic stream open between dictations: recording starts instantly, and up to half a second captured before you press the hotkey is included, so first words are never clipped.

  • macOS shows the microphone indicator while Dictly runs — that's the honest cost. Turn the option off in Settings if you prefer.
  • Bluetooth microphones are excluded (holding a capture stream open would pin them to low-quality call audio); they keep the on-demand start.

Dictly 1.1

Choose a tag to compare

@vlr-code vlr-code released this 02 Jul 19:05

Language, microphone and hotkey fixes — the two most-reported issues addressed.

Language (#1, #4)

  • The default spoken language now follows your Mac's system language instead of being hard-coded (#4). First-run onboarding also asks for it explicitly.
  • New: switch between two languages with a dedicated hotkey (#1). Enable it in Settings → “Switch between two languages with a hotkey”; the active language (flag or “auto”) shows next to the menu-bar icon and applies to your next dictation.

Microphone & audio

  • New Microphone picker in the menu bar. Picking a mic sets it as the Mac's default input, and the recorder follows.
  • Fixed repeated recordings capturing no audio on Bluetooth / Continuity / virtual microphones.

Hotkeys & permissions

  • Fixed a bug where a second hotkey could never fire.
  • The push-to-talk and language-switch hotkeys can't be assigned to the same key.
  • Clearer Accessibility guidance: an actionable alert when auto-paste is blocked, a warning at launch when a modifier-only hotkey (Fn, right ⌥) can't work without the grant — and hotkeys now recover automatically the moment access is granted, no relaunch needed.
  • The onboarding window stays fully on screen on small displays.

Known limitation: the bundled model is still Whisper base (~139 MB), which is weak for non-English speech. For best accuracy, download a larger model right after install in Settings → Transcription models — recommended: Large v3 · compressed (547 MB, one-time download).

Dictly 1.0

Choose a tag to compare

@vlr-code vlr-code released this 14 May 09:36

First public release.

  • Whisper base multilingual bundled (~139 MB) — works offline from first launch
  • For best accuracy, switch to a larger model right after install in
    Settings → Transcription models. Recommended: Large v3 · compressed
    (547 MB, one-time download from HuggingFace) — same quality as full
    Large v3 at a third of the size. RTF ≈ 0.15 on Apple Silicon (a 10-second
    utterance transcribes in ~1.5 seconds).
  • 99 languages, fully on-device
  • Push-to-talk hotkey (default: Fn)
  • Auto-paste into the focused app via Accessibility