v1.3.0

What's new

• A global listening toggle hotkey - configurable via Settings → General.
• Hold to talk - an alternative mode, for those who don't want to use the always-on default.
• Matching inserted text into the context - spacing and sentence casing match what's already around the cursor.

Fixes

• First-transcription is no longer delayed with Whisper models.
• Error reporting when the engine couldn't load.
• The setup window going to the background after a permission grant.

Full Changelog: v1.2.1...v1.3.0

v1.2.1

Fixed

Setup could end with an error on a brand-new install.

If you granted the microphone and accessibility permissions while the
transcription model was still downloading, Overhear would close the setup
window and then show a red error icon. Dictation never became available,
and quitting and reopening the app was the only way to fix the issue.

Full Changelog: v1.2.0...v1.2.1

v1.2.0

What's new

• Runs on Neural Engine. - Transcription and wake word detection now run natively inside the app, and are no longer dependent on Python. Transcription runs on the Neural Engine through CoreML.
• Multiple Whisper models supported. - A new settings pane lets you choose which model is used for transcription.
• Parakeet as an alternative to Whisper. - NVIDIA's Parakeet is available too, either English-only or across twenty-four languages. It is faster than Whisper, provides better accuracy, and run on the Neural Engine.
• UX cleanup. - The settings menu has been revamped for better usability. The menu bar now shows when a model is loading and when setup still needs you. Dictation carries across a change of input device instead of stopping.

Fixes

• Start Listening did nothing after an input device changed, until the app was restarted.
• Reopening the microphone could abort the whole app on a device reconfigure.
• Speaking a selected language sometimes produced the output translated to English. The output now stays inside the languages you picked.

Installation

Download Overhear.dmg below, open it, and drag Overhear to Applications.

Full Changelog: v1.1.1...v1.2.0

v1.1.1

What's fixed

• Installing XCode Developer Tools. - The Overhear setup process now actively waits until it sees python available, and then automatically continues the dependency installation process.
• Permission labels cleanup. - Removed unnecessary labels in the "Grant Permissions" window.

Installation

1. Download Overhear.dmg below, open it, and drag Overhear to Applications.
2. Clear the quarantine flag (required for unsigned apps): /usr/bin/xattr -cr /Applications/Overhear.app
   listening.

Full Changelog: v1.1.0...v1.1.1

v1.1.0

What's new

• New menu bar icon. - The static icon is replaced by a live mark that shows what the app is doing: dim when idle, solid while listening, orange while it hears you or transcribes, and red when the cancel word fires. The overlay uses the same mark, so both read the same at a glance.
• New app icon, - drawn from the same live mark and rendered in code at every size (scripts/make-icon.sh).
• Clearer labels. - The menu item is now Start Listening / Stop Listening, and Settings reads Start listening on launch and Show overlay window while

Installation

1. Download Overhear.dmg below, open it, and drag Overhear to Applications.
2. Clear the quarantine flag (required for unsigned apps): /usr/bin/xattr -cr /Applications/Overhear.app
   listening.

Full Changelog: v1.0.0...v1.1.0

v1.0.0

Overhear is a menu-bar dictation app for macOS. It listens continuously and pastes transcriptions into whatever text field is focused - Slack, your editor, a browser form, anything that accepts a paste. Speech recognition runs entirely on your Mac using faster-whisper and openwakeword; no audio ever leaves your machine.

Highlights

• Continuous, hands-free dictation - launch the app and start talking. No hotkeys, no button to hold. Text is transcribed in batches as you pause and pasted automatically.
• Works everywhere - transcriptions are delivered via the clipboard and a simulated Cmd+V, so they land in any app that accepts a paste, including ones with no native dictation support.
• Multi-language - select the languages you use in Settings and Overhear detects which one you're speaking on the fly, no manual switching needed.
• Cancel word - say "Alexa" (or choose another wake word) to discard what you just said and keep dictating. A shake animation in the menu bar and overlay confirms the cancel.
• Custom hot words - install your own openwakeword .onnx models to use as the cancel word.
• Privacy-first - all models are downloaded once during first launch and everything runs offline after that.
• Automatic setup - on first launch, the app creates its own Python environment, downloads the speech models, and guides you through granting Microphone and Accessibility permissions.

Installation

1. Download Overhear.dmg below, open it, and drag Overhear to Applications.
2. Clear the quarantine flag (required for unsigned apps): bash xattr -cr /Applications/Overhear.app