Speech

A macOS menu bar app for speech-to-text using local Whisper models. All transcription happens on-device — no cloud services, no API keys, complete privacy.

Features

• Hold-to-record: Hold the hotkey to record, release to transcribe and auto-paste
• Local processing: Whisper models running entirely on your Mac
• Privacy-first: No data leaves your device
• Multiple engines: Whisper, Moonshine, Parakeet, SenseVoice
• Model profiles: Save different model configurations and switch between them
• Multi-language: 25+ languages with auto-detection
• Customizable hotkey: Default is ⌥Space (Option + Space)
• Auto-update: Checks for updates from GitHub releases

Tech Stack

• Tauri 2 — native macOS app shell
• Rust — audio recording, transcription, hotkeys, paste injection
• Svelte 5 + Tailwind CSS — UI (menu bar panel, overlays, settings)
• transcribe-rs — local Whisper inference

Installation

Requirements

• macOS 14.0 or later

Download & Install

1. Download the latest release from Releases
2. Unzip and drag Speech.app to /Applications
3. Open Speech.app

macOS Gatekeeper: Since the app is not notarized, macOS may block it. Right-click → Open → Open again, or go to System Settings → Privacy & Security → Open Anyway.

Build from Source

git clone https://github.com/kennetkusk/speech.git
cd speech
bun install
./build_app.sh
cp -R .build/Speech.app /Applications/
open /Applications/Speech.app

Setup

On first launch, grant these permissions (check status in Settings → Permissions):

1. Microphone — required for recording
2. Accessibility — required for auto-paste and global hotkey
3. Input Monitoring — required for keyboard simulation

Then download a model in Settings → Models.

Usage

1. Hold ⌥Space to start recording
2. Speak clearly
3. Release to transcribe and auto-paste

Press Escape while recording to cancel.

Settings

Click the menu bar icon → Settings to configure:

• General: Hotkey, language, auto-paste, launch at login
• Models: Download/manage transcription models
• Profiles: Save model configurations for quick switching
• Permissions: Check macOS permission status

Privacy

All audio is processed locally. No data is sent to external servers.

License

MIT — see LICENSE.