Repository navigation
Releases: DanielOu1208/aerivoice
Release list
AeriVoice 0.2.5 — Chosen microphones and pasting into Mail
AeriVoice 0.2.5 fixes dictation failing at once on a microphone chosen in Settings, pastes into a new Mail message instead of copying, and says so when cleanup failed.
What's fixed
- A chosen microphone starts: if the microphone picked under Settings → Dictation → Microphone had a different sample rate or channel count from the Mac's default devices, as a USB wireless receiver can, every dictation ended at once with an error. AeriVoice now opens the microphone in its own format.
- Fallback to the default microphone: if a chosen microphone is connected but still cannot be started, the dictation runs on the system default instead and the notch shows "Using the default microphone". See Microphone.
- Paste into a new Mail message: dictating into Mail's compose body copied the text instead of pasting it. It now pastes.
- Cleanup failures stay visible: when cleanup fails and the raw transcript is pasted, the notch shows "Cleanup failed—used raw text" for two seconds. Before, the check mark replaced it almost immediately.
- Clipboard restored sooner in an empty Notes or Messages field: a paste into an empty field is now verified, so your clipboard comes back once the text has landed instead of after the fallback delay.
Install or upgrade
Requires Apple Silicon and macOS 26 or newer. Use Settings → General → Check for Updates, or download the DMG below, open it, and drag AeriVoice into Applications. Existing settings and Keychain credentials are retained.
The DMG is Developer ID signed, notarized by Apple, and stapled. A SHA-256 checksum and debugging symbols are attached.
AeriVoice 0.2.4 — Spoken language and steadier dictation
AeriVoice 0.2.4 lets you choose the language you speak, keeps a dictation going when macOS is slow to hand over your API keys, and removes several waits and background chores across transcription providers.
What's new
- Spoken language: choose a language under Settings → Dictation → Transcription, or keep Auto-detect, the default, which works as before. Soniox restricts recognition to the chosen language. Grok and Meta use it as a hint and can still follow another language. Cartesia detects its languages itself, Nemotron transcribes English, and Apple Speech keeps its own language setting. See Providers and models.
What's fixed
- A slow Keychain no longer loses the dictation: if macOS took several seconds to return your API keys, the audio recorded meanwhile could overflow and the dictation failed. That audio is now kept and sent once the connection opens.
- No keyboard freeze while keys load: API keys are read off the main thread, once per dictation. Releasing the shortcut during a slow read cancels at once instead of stalling every key press until the read finished.
- Local Nemotron ready after wake: a model that is already loaded stays ready when you wake the Mac or change settings, instead of re-checking all of its files and refusing to dictate meanwhile. It also skips rebuilding your Dictionary hints when they haven't changed.
- Apple Speech ready after wake: re-checking the language you already use no longer marks it not ready.
- Grok keeps its prepared connection when a Dictionary edit doesn't change the terms Grok receives.
Smaller improvements
- The clipboard check after a paste runs off the main thread.
- The last moment of audio after you release the shortcut goes out in one piece, which helps Meta and Grok finish sooner.
- Lighter background work: diagnostics housekeeping, model download progress, and API key status refreshes do less.
Install or upgrade
Requires Apple Silicon and macOS 26 or newer. Use Settings → General → Check for Updates, or download the DMG below, open it, and drag AeriVoice into Applications. Existing settings and Keychain credentials are retained.
The DMG is Developer ID signed, notarized by Apple, and stapled. A SHA-256 checksum and debugging symbols are attached.
AeriVoice 0.2.3 — Cartesia transcription and tighter recording edges
AeriVoice 0.2.3 adds Cartesia as a transcription provider, keeps everything you say up to the moment you release the shortcut, and starts the microphone sooner when sounds are off.
What's new
- Cartesia transcription: choose Cartesia Ink 2 or Cartesia Ink Preview in Settings → Dictation and connect a Cartesia API key. Ink 2 is Cartesia's stable model; Ink Preview is its preview model and can change without notice. Both transcribe English, French, Hindi, Japanese, and Spanish only. Your Dictionary is sent as hints, up to 100 terms totaling 1,200 characters (fewer in non-Latin scripts). Terms that don't fit stay saved and are listed in Dictation settings. See Providers and models.
- Recording ends exactly at release: audio reaches AeriVoice in blocks of about a tenth of a second, and the block in progress when you released the shortcut could be dropped, cutting off the end of the last word. AeriVoice now waits for that block and keeps everything up to the release. Recording no longer continues for 50 ms afterwards.
- Faster start with sounds off: the microphone starts at the key press instead of after AeriVoice's key and permission checks. After you cancel a dictation, the next one starts on a microphone engine that is already prepared.
- Recommended model: the Dictation model picker marks Soniox as the recommended cloud model.
What's fixed
- Microphone lost while starting: if your input device went away in the first moments of a dictation, AeriVoice could record silence and report "No speech detected". It now reports the microphone error.
Install or upgrade
Requires Apple Silicon and macOS 26 or newer. Use Settings → General → Check for Updates, or download the DMG below, open it, and drag AeriVoice into Applications. Existing settings and Keychain credentials are retained.
The DMG is Developer ID signed, notarized by Apple, and stapled. A SHA-256 checksum and debugging symbols are attached.
AeriVoice 0.2.2 — Paste and update fixes
AeriVoice 0.2.2 fixes pasting into deeply nested web apps, restores your clipboard after pasting into terminals, and stops update checks from freezing the app.
What's fixed
- Pasting into web apps like Claude.ai: before pasting, AeriVoice checks every container around the text field to make sure it isn't a password field. That check gave up after 32 levels, and some web apps nest their text box deeper (Claude.ai in a Chromium-based browser sits 33 levels down), so the result was copied instead of pasted. The limit is now 128 levels, and every level is still checked.
- Clipboard restore in terminals: when an app redraws its screen instead of showing the exact pasted text (for example Claude Code in Ghostty), AeriVoice couldn't confirm the paste and never restored your previous clipboard. It now restores it after the fallback delay set in Settings → General → Clipboard. If the field didn't change at all, the dictation stays copied so you can paste it manually.
- Check for Updates no longer freezes AeriVoice: while the "You're up to date" message was open, dictation stopped responding, and quitting could hang. Update messages now open without blocking the rest of the app.
- Crash with performance logging on: AeriVoice could crash after five idle minutes when performance logging was enabled.
Install or upgrade
Requires Apple Silicon and macOS 26 or newer. Use Settings → General → Check for Updates, or download the DMG below, open it, and drag AeriVoice into Applications. Existing settings and Keychain credentials are retained.
The DMG is Developer ID signed, notarized by Apple, and stapled. A SHA-256 checksum and debugging symbols are attached.
AeriVoice 0.2.1 — Smoother notch, cleaner microphone capture, and steadier connections
AeriVoice 0.2.1 makes dictation feel smoother and loses fewer words at the start and end of a recording. It is the first release without the beta label.
What's new
- Smoother notch: the recording notch animates without resizing its window every frame. In our 30-cycle test, detected hitches fell from 95 to 54 and the longest from 33 to 17 ms.
- Recording starts sooner: with sounds on, the microphone starts while the start sound plays, so speech is kept from about 200 ms after the key press instead of about 475 ms.
- Fewer clipped endings: capture continues briefly after you stop, and the stop sound plays only after the microphone closes.
- Microphone choice: pick an input in Settings → Dictation → Microphone. System default stays the default; a disconnected choice falls back to it.
- Device changes mid-dictation: switching inputs no longer silences the rest of a recording. If the new input cannot start, AeriVoice finishes with what it already captured.
- Clearer input: multichannel microphones are mixed down instead of using only the first channel, and a light filter removes low rumble such as fans and desk bumps.
- Clipboard restore delay: Settings → General → Clipboard sets how long to wait before restoring your clipboard when an app cannot confirm the paste.
- Faster first requests: Groq and OpenRouter cleanup connections are prepared when you start dictating, and a prepared Grok connection is kept for up to five minutes.
- Retiring OpenRouter models: retired models are hidden, retiring ones are labeled, and Settings warns if your selected model is retiring.
- Reliability: literal
<think>text is preserved with Cerebras, incomplete provider responses fall back to your raw transcript, and pending clipboard restoration is handled reliably around update restarts.
Timing figures come from local tests on one Mac and are not guarantees for every device or network. See the optimization notes for methods and limitations.
Install or upgrade
Requires Apple Silicon and macOS 26 or newer. On 0.2.0-beta.1, use Settings → General → Check for Updates. Otherwise download the DMG below, open it, and drag AeriVoice into Applications. Existing settings and Keychain credentials are retained.
The DMG is Developer ID signed, notarized by Apple, and stapled. A SHA-256 checksum and debugging symbols are attached.
AeriVoice v0.2.0-beta.1 — In-app updates, Stats, and a new installer
AeriVoice 0.2.0-beta.1 adds built-in updates, local usage Stats, refined Settings, and a new installation window.
What's new
- In-app updates: daily checks, optional alerts, and installation when you choose. The update feed and downloads are signed; Offline mode blocks updater networking.
- Local Stats: word counts, dictations, recording time, and speaking pace. Daily totals stay on your Mac, without transcripts, and can be disabled or cleared.
- Refined Settings: native navigation, colorful sidebar icons, and clearer provider choices.
- Branded installer: a light, Retina installation window with drag-to-Applications setup.
- Reliability: bounded Stats charts and coordination between update restarts, active dictation, clipboard restoration, and pending Stats saves.
Install or upgrade
Requires Apple Silicon and macOS 26 or newer. Download the DMG below, open it, and drag AeriVoice into Applications.
If your current version has no built-in updater, quit AeriVoice and replace it manually this time. Existing settings and Keychain credentials are retained. Future updates are available through Settings → General → Check for Updates.
This is a beta release. The DMG is Developer ID signed, notarized by Apple, and stapled. A SHA-256 checksum and debugging symbols are attached.
AeriVoice v0.1.0-beta.10 — Grok transcription and clearer provider icons
AeriVoice v0.1.0-beta.10 — Grok transcription and clearer provider icons
- Grok transcription: connect an xAI API key in Providers, then choose Grok Voice Transcribe 2.0 in Dictation.
- Dictionary hints: send up to 100 eligible terms to Grok. Unsupported entries stay saved, with exclusions visible in Settings; recognition hints do not guarantee exact spelling.
- Faster Grok startup and catch-up: reuse a prepared connection for up to 30 seconds and drain buffered audio at up to 1.35× real time without changing the audio samples. In our small production catch-up comparison, median cold-stop delay fell from 585 to 342 ms; this is not a guarantee for every recording or network.
- More reliable connection handling: fresh and prepared Grok sessions share one transport implementation, with regression coverage for duplicate finish requests, cancellation, and replacement sessions.
- Clearer provider icons: corrected SpaceX clearspace, OpenRouter contrast, and Cerebras geometry in Settings and the README.
Grok requires xAI API credits. Mixed-language recognition and AI cleanup can still make mistakes; review important text. Catch-up pacing has been tested empirically but is not an explicit xAI support guarantee. See Grok validation for benchmark methods and limitations.
Requires Apple Silicon and macOS 26+. Updates are manual. Existing settings and saved credentials are retained.
Updating
Download AeriVoice-v0.1.0-beta.10-arm64.dmg and its matching .sha256 file. Verify the checksum, open the DMG, quit AeriVoice, and replace the app in Applications.
Build: 2026092002. Debug symbols and a source/build receipt accompany the installer.
AeriVoice v0.1.0-beta.9 — Cleanup styles and a new icon
AeriVoice v0.1.0-beta.9 — Cleanup styles and a new icon
- Compose (Experimental): format dictated lists and paragraphs, apply clear spoken corrections, and refine your writing within the current recording.
- Custom (Experimental): choose Custom in the cleanup-style dropdown to reveal your instructions. Set tone, spelling, translation, or formatting on top of Polished cleanup. Instructions apply only in Custom and stay saved when you switch styles. Empty Custom uses Polished cleanup.
- Clearer style choices: Faithful, Polished, Compose, and Custom each have a short description. Polished remains the default, with improved cleanup of fillers and phrasing.
- New microphone mark: refreshed app, menu-bar, settings-sidebar, and README icons. The menu-bar mark turns orange while recording.
Compose and Custom can miss edits or change meaning. Review important text. Higher reasoning may help where supported, but adds latency and has not been verified to improve reliability in our benchmarks. Your reasoning setting is unchanged. Offline mode skips all AI cleanup.
Requires Apple Silicon and macOS 26+. Updates are manual.
Updating
Installer: AeriVoice-v0.1.0-beta.9-arm64.dmg, build 2026091701, with a matching .sha256 file and debug symbols. Validation and remaining acceptance checks are tracked in the pre-publication validation record.
Final CI and installer checks, including the initial CI failure and follow-up results, are recorded in the attached release-validation.txt.
AeriVoice v0.1.0-beta.8 — Local dictation and Offline mode
- Dictate on your Mac: choose Apple Speech or the recommended NVIDIA Nemotron model for English. Apple Speech can reuse installed language support; Nemotron requires a 611 MB download. Local accuracy may be lower than cloud models.
- Offline mode: enable it from General or the menu bar once a local model is set up. It bypasses AI cleanup and requires no cloud keys. Your normal provider choices are preserved when switching back online.
- Simpler setup: local models are prominent in onboarding. One model selector covers cloud and local options, while Providers owns downloads, language setup, status, and removal. Interrupted NVIDIA downloads can be resumed or removed after reopening the app.
- More shortcut control: choose Hybrid, Toggle, or Hold, and optionally distinguish left and right Command and Option keys. Trackpad activation is not included.
- Settings and recovery fixes: a larger minimum settings window, clearer download failures, safer Apple language reservations, and onboarding recovery when a model or permission becomes unavailable.
Requires Apple Silicon and macOS 26+. Selecting a local model alone does not disable cloud cleanup; enable Offline mode to keep dictation entirely local. macOS manages Apple language assets and may maintain or retry previously requested downloads independently.
Updating
Installer: AeriVoice-v0.1.0-beta.8-arm64.dmg, build 2026091405. The matching .sha256 file and debug symbols are packaged alongside it. Existing keys and preferences are intended to remain intact; clean-install and upgrade acceptance are tracked in the readiness checklist.
AeriVoice v0.1.0-beta.7 — More cleanup models and native settings
AeriVoice v0.1.0-beta.7 — More cleanup models and native settings
- More OpenRouter cleanup models: start with recommended presets, then browse and search compatible text models or enter a model ID. Model capabilities and reasoning options refresh from OpenRouter and are cached for offline use.
- Native macOS settings: a permanent sidebar, compact provider rows with icons, simpler headings, and a cleaner model picker. Dictionary entry is left-aligned with Add on the right.
- Flexible onboarding: connect one transcription provider and one cleanup provider. OpenRouter is optional; direct Cerebras is supported as a standard option, while Groq remains experimental.
- Polished by default: applies when no cleanup style is saved, including existing installations. Explicitly saved styles are preserved.
- Settings fixes: scrolling content stays below the toolbar title. Closing Settings dismisses open sheets and cancels credential verification, preventing late responses from saving a dismissed replacement.
OpenRouter is still used for cleanup only, not transcription.
Updating
Quit AeriVoice, download the DMG and checksum, and verify:
shasum -a 256 -c AeriVoice-v0.1.0-beta.7-arm64.dmg.sha256Open the DMG, replace AeriVoice in Applications, and reopen it. Existing keys and saved settings are retained. Requires Apple Silicon and macOS 26+.
Validation
Normal and distribution suites each passed 343 tests, with three opt-in live benchmarks skipped. The installer is signed, notarized, and Gatekeeper-verified. Remaining fresh-install, live-provider, and device checks are listed in the readiness record.