Menu-Whisper

A macOS menu bar application that provides offline speech-to-text transcription using Whisper-family models and automatically inserts the transcribed text into the currently focused application.

Overview

Menu-Whisper is designed to be a privacy-focused, offline-first speech recognition tool for macOS. It runs entirely locally on Apple Silicon machines, requiring no internet connection during normal operation (only for initial model downloads).

Key Features

• 100% Offline Operation: Audio and text never leave your device
• Apple Silicon Optimized: Built specifically for M1/M2/M3 processors with Metal acceleration
• Global Hotkey Support: Default ⌘⇧V (configurable)
• Smart Text Insertion: Clipboard paste with typing fallback
• Secure Input Detection: Respects password fields and secure contexts
• Multiple Models: Support for various Whisper model sizes and variants
• Multilingual: Spanish and English interface and recognition

Requirements

• macOS: 13.0 (Ventura) or later
• Hardware: Apple Silicon (M1, M2, or M3 processor) - Intel Macs are not supported
• Xcode: 15.0+ for building from source
• Permissions: Microphone, Accessibility, and Input Monitoring access

Build Requirements

Development Environment

• macOS 13+ with Xcode 15.0+
• Swift 5.9+
• Swift Package Manager (included with Xcode)

System Dependencies

• AVFoundation framework (audio capture)
• Carbon framework (global hotkeys)
• AppKit/SwiftUI (UI components)

Third-party Dependencies

• whisper.cpp (C/C++ library for speech recognition with Metal support)

Installation

Note: This project is currently in development. Pre-built binaries will be available as signed and notarized .dmg files once complete.

Building from Source

1. Clone the repository:

   git clone <repository-url>
   cd tellme
   

2. Open the project in Xcode or use Swift Package Manager:

   swift build -c release
   

3. For development, open Package.swift in Xcode.

Architecture

The application is structured with modular components:

• App: SwiftUI interface with AppKit bridges
• Core/Audio: AVAudioEngine capture and processing
• Core/STT: Speech-to-text engines (whisper.cpp, future Core ML)
• Core/Models: Model management and downloads
• Core/Injection: Text insertion with secure input handling
• Core/Permissions: System permission management
• Core/Settings: User preferences and configuration

Privacy & Security

• No Telemetry: Zero data collection or remote analytics
• Local Processing: All audio processing happens on-device
• Secure Input Respect: Automatically detects and respects secure input contexts
• Permission-Based: Requires explicit user consent for system access

Development Status

This project is currently in active development following a phased approach:

• Phase 0: Project scaffolding ⬅️ Current
• Phase 1: Hotkey + HUD + Audio capture
• Phase 2: STT integration with whisper.cpp
• Phase 3: Text insertion system
• Phase 4: Preferences and UX polish
• Phase 5: Distribution and packaging

See TODO.md for detailed development progress and TECHSPEC.md for complete technical specifications.

License

MIT License - see LICENSE for details.

Contributing

This project follows a structured development approach with clear phases and acceptance criteria. Please refer to the technical specification and TODO list before contributing.