VoiceType 2.4.1

Hotfix release for the v2.4.0 launch crash caused by packaged localization resources not being found in the installed app.

• Fix localization bundle lookup in production app bundles.
• Package and verify all SwiftPM resource bundles.
• Validate the signed app, mounted DMG, and Sparkle ZIP with a production launch probe during releases.

If VoiceType 2.4.0 crashes before Sparkle can update it, download the DMG below and replace the existing app in Applications.

VoiceType 2.4.0

Your language, end to end

VoiceType is now substantially more multilingual from first launch through dictation. The app, cleanup, model selection, and documentation all work together so dictation feels native in more languages while keeping your voice on your Mac.

Highlights

• Dictate in 30+ languages with language-aware cleanup, including punctuation and filler-word handling tailored to each language.
• VoiceType’s interface is localized in 16 languages, and the models page only offers engines that support your selected language.
• Added language packs for German, Spanish, French, Italian, Japanese, Korean, Dutch, Polish, Brazilian Portuguese, Russian, Swedish, Turkish, Ukrainian, Vietnamese, and Simplified Chinese.
• The README is now available in all 15 of those languages, with privacy and on-device operation front and center.
• The dictation indicator stays out of the way at rest and appears only while listening.

Private by design

Audio and transcripts remain on your Mac. This release continues to use on-device transcription and cleanup by default, with no account or telemetry required.

VoiceType 2.3.1

A way back to a dismissed update

Dismissing the update dialog used to be a dead end — the prompt only returned on the next daily background check, and a skipped version never came back on its own.

Now, whenever an update is pending, an Update available row appears in the sidebar just above Setup. Click it to reopen the update dialog anytime — even for a version you previously skipped. The row disappears once you're up to date.

VoiceType 2.3.0

Dictionary

Word replacements are now a first-class Dictionary page in the sidebar:

• Composer — add "when you say → VoiceType types" pairs with Return-to-add and inline duplicate detection.
• Editable list — edit pairs in place, delete on hover, duplicates flagged, search appears once the list grows past 8 entries.
• Live playground — type a sentence the way the transcriber might hear it and watch your dictionary rewrite it in real time, replaced words highlighted. Mirrors the pipeline's matching exactly (whole words, any casing, top to bottom).
• Starter examples for an empty dictionary.

Also since 2.2.0

• Built-in rules are now the default cleanup engine — instant, deterministic, symbol-aware (spoken symbols like "comma" and "new line" render correctly).
• Word replacements apply after any cleanup engine, cloud or local.
• Language selection is part of the guided setup flow, and the Settings language is enforced everywhere — no inference, no drift.
• Cleanup evaluation harness can benchmark the deterministic path (--engine rules).

VoiceType 2.2.0

Smarter, more faithful dictation cleanup

The on-device AI cleanup pass was overhauled and validated against the live Apple Intelligence model with a new eval harness (6 batteries, 456 outputs):

• No more shortened dictations. The cleanup prompt now leads with a hard rule against summarizing, and a deterministic guard discards any model output that loses content (too short, fabricated, or missing your opening words) — falling back to safe rule-based cleanup instead.
• App-aware cleanup. VoiceType now notices which app you're dictating into: terminals get shell-command treatment (git commit -m "…", --verbose, lowercase, no trailing period), code editors bias toward identifier rendering (main.py, blue_file, parseRequest), prose apps stay prose.
• Consistent finishing touches. Sentence capitals, the pronoun "I", and question marks are now guaranteed deterministically rather than left to the model.
• Settings: toggle to hide the resting indicator pill.

Everything stays on-device; nothing leaves your Mac.

VoiceType 2.1.0

Dictation that survives your headphones

AirPods can't kill your dictation anymore. Audio capture now rides AVCaptureSession instead of AVAudioEngine, so connecting or switching headphones mid-dictation no longer cuts the recording short — route changes are absorbed instead of tearing down the mic.

Sound cues you can trust. The start/stop chimes now play reliably every time — including the start cue on Bluetooth headphones, which used to get swallowed while the audio route was switching over.

Everything still runs on-device — your audio never leaves your Mac.

Download VoiceType.dmg below (signed + notarized — just drag to Applications). Existing installs update automatically via Sparkle.

VoiceType 2.0.0

Speak in your language, and a new NVIDIA engine

Language that actually sticks. Pick your language in Settings and VoiceType now transcribes and cleans up in it — end to end. The speech models are told which language to expect (no more English-by-default guessing), and the cleanup pass keeps your words in that language instead of drifting to English.

New engine — NVIDIA Nemotron 3.5 ASR (0.6B). A fresh multilingual on-device model with broad language coverage, downloadable from the Models page and run entirely on your Mac's Neural Engine. It uses your selected language as a hint and gracefully auto-detects when your speech doesn't match, so it never comes back empty.

Redesigned Settings. General and Cleanup are now one clean scrolling page with a proper sidebar — including the new Language picker.

See what you've downloaded. A new button on the Models page opens a single Finder folder with all your downloaded speech models in one place.

Everything still runs on-device — your audio never leaves your Mac.

Download VoiceType.dmg below (signed + notarized — just drag to Applications). Existing installs update automatically via Sparkle.

VoiceType 1.0.2

Smarter dictation cleanup

The on-device (Apple Intelligence) and Groq cleanup passes now post-process your speech more intelligently:

• Compact code rendering — when you're clearly dictating code, spoken file names, symbols, and identifiers come out as real code instead of separate words: "open parser dot pie" → open parser.py, "print open paren x comma y close paren" → print(x, y), "get underscore user data" → get_user_data. Ordinary prose is left alone ("the dot product" stays prose).
• Faithful self-correction — false starts resolve to what you meant ("I want two, no three" → "I want three") while your word order and word choice are otherwise preserved. Cleanup no longer reorders or restructures your sentences.

Download VoiceType.dmg below (signed + notarized — just drag to Applications). Existing installs update automatically via Sparkle.

VoiceType 1.0.1

A small housekeeping release on top of 1.0.0. No changes to app behavior.

Changes

• 📄 The project is now MIT licensed — added a LICENSE file and surfaced it in the README.
• 🛠️ Release CI is now a manual (workflow_dispatch) fallback, so tag pushes no longer trigger a failing run. Notarized releases continue to be cut locally.

Install

Download VoiceType.dmg, drag it to Applications, and open it. Signed and notarized by Apple — a plain double-click works. Requires macOS 26+ (Apple Silicon). Existing installs will update themselves automatically via Sparkle.

VoiceType 1.0.0

VoiceType 1.0 — speak anywhere, get clean text instantly, on-device.

The first stable release. A fast, private voice-dictation app for macOS: hold a key, talk, and your words land as clean, punctuated text in whatever app you're using. Audio and transcripts stay on your Mac unless you opt into a cloud engine.

Highlights

• 🎙️ Press-to-talk dictation anywhere via a global hotkey (default Right Option), with a live frosted waveform pill.
• 🔒 Private by default — on-device transcription (Apple SpeechTranscriber) and cleanup (Apple Intelligence), with a local Whisper fallback and an opt-in Groq cloud upgrade.
• 📊 Home dashboard — your words, pace, and day streaks, a full activity heatmap, where you dictate, and a friendly on-device usage summary with a cute border-trace animation. The summary uses aggregate counts only — never your transcript text.
• ✨ Smart cleanup — punctuation, capitalization, and filler removal without changing your words.
• 🧩 Guided setup with on-page permission + hotkey controls.

Install

Download VoiceType.dmg, drag it to Applications, and open it. This build is signed and notarized by Apple, so it launches with a normal double-click. Requires macOS 26+ (Apple Silicon).

Updates are automatic from here on via Sparkle — every update is cryptographically signed and verified.