Mac Speech to AI to Text

A macOS menu bar app that turns speech into refined, ready-to-send text anywhere you type — powered by local Whisper transcription and AI cleanup.

What It Does

Trigger recording with a global shortcut or wake phrase, speak naturally, and have your words transcribed locally, refined with AI, and inserted directly into the active text field — whether that's a chat window, browser, editor, or AI tool.

No app switching. No copy-pasting. Just speak and send.

Features

• Global shortcut & wake phrase activation — trigger from anywhere on your Mac
• Local Whisper transcription — privacy-first, on-device speech-to-text via WhisperKit
• AI cleanup — refine raw dictation into polished text with configurable prompts
• Flexible AI providers — OpenAI, Anthropic, or custom endpoints with selectable models
• Direct text insertion — results go straight into the active text field
• Spoken send phrase — hands-free recording control
• Auto media pause/resume — pauses Spotify/Apple Music while recording
• Recording history — browse, replay, and reuse past transcriptions
• Custom prompt library — save and switch between cleanup prompts
• Multiple shortcut bindings — assign different shortcuts to different prompts
• Notification sounds — configurable beep presets and volume
• Menu bar app — always available, never in the way

How It Works

1. Press your shortcut (or say your wake phrase)
2. Background media pauses automatically
3. Speak your message
4. Say your send phrase or press the shortcut again to stop
5. Audio is transcribed locally with Whisper
6. AI cleanup refines the transcription (optional)
7. Text is inserted into the active text field or copied to clipboard

Requirements

• macOS 14.0 (Sonoma) or later
• Microphone access
• Accessibility permission (for text field detection and insertion)
• Input Monitoring permission (for global keyboard shortcuts)

Install

Pre-built App

1. Download Mac-Speech-to-AI-to-Text.zip from the releases/ folder
2. Unzip and drag Mac Speech to AI to Text.app to your Applications folder
3. Open the app — grant Microphone, Accessibility, and Input Monitoring permissions when prompted

Build From Source

git clone https://github.com/savedpixel/mac-speech-to-ai-to-text.git
cd mac-speech-to-ai-to-text
swift build
bash scripts/build-app.sh
open "build/Mac Speech to AI to Text.app"

Or open Package.swift in Xcode and run directly.

Stack

Layer	Technology
Language	Swift 5.9+
UI	SwiftUI + AppKit (NSStatusItem)
Audio	AVAudioEngine
Transcription	WhisperKit (local Whisper inference)
Input	CGEvent (shortcuts, key simulation), Accessibility API
Speech	Speech framework (wake/send phrase detection)
Persistence	UserDefaults / @AppStorage
Logging	OSLog

Project Structure

MacSpeechToAIToText/
  App/            — Entry point, AppDelegate
  Audio/          — Recording, playback, media control, signal sounds
  Core/           — Settings, pipeline coordinator, history, AI providers
  Input/          — Shortcuts, text insertion, wake/send phrase listeners
  Transcription/  — Whisper engine, AI cleanup
  UI/             — Menu bar, preferences, recording overlay, history

Configuration

All settings are accessible from the app's main window:

• Shortcuts — Set global hotkeys, bind prompts to specific shortcuts
• AI Cleanup — Choose provider, model, and endpoint; toggle cleanup on/off
• Prompts — Create and manage cleanup prompt templates
• Audio — Select microphone, configure notification sounds
• Send Phrase — Set spoken phrase to stop recording hands-free
• Wake Phrase — Set spoken phrase to start recording hands-free
• History — Auto-delete old recordings, browse past transcriptions

Privacy

• Core transcription runs entirely on-device using Whisper — no audio leaves your Mac
• AI cleanup (optional) sends only the transcribed text to your chosen provider
• No telemetry, no analytics, no accounts

License

MIT