SpeakType

Fast Voice-to-Text for macOS

Press a hotkey, speak, and instantly paste text anywhere on your Mac.

---

What is SpeakType?

SpeakType is a privacy-first voice dictation tool for macOS. Audio transcription runs locally using OpenAI's Whisper AI model via WhisperKit, and optional text enrichment can run either through a local Ollama model or through OpenAI. Audio never leaves your Mac; when cloud enrichment is enabled, only transcript text is sent.

• Privacy First - Audio stays local, with explicit Local vs Cloud enrichment modes
• Lightning Fast - Optimized for Apple Silicon
• Works Everywhere - Any app, any text field
• Open Source - Audit every line of code yourself

Enrichment Modes

• Raw - Paste the transcript as-is
• Clean - Fix punctuation, casing, spacing, and obvious grammar
• Polish - Improve readability while preserving intent
• Email - Rewrite the dictation as a ready-to-send email body

Enrichment Providers

• OpenAI (Cloud) - Uses the OpenAI Responses API with your own API key
• Ollama (Local) - Uses a local Ollama model running at http://localhost:11434

Ollama Model Manager

• Browse installed local Ollama models from Settings → AI
• Download curated local models in-app, including Gemma 4, Qwen 3, and Llama 3.2 options
• See exact installed disk usage plus family, parameter size, quantization, and context metadata
• Run on-demand speed tests to measure tokens/sec on your Mac

---

Installation

Requirements

• macOS 13.0+ (Ventura or newer)
• Apple Silicon (M1+) recommended
• 2GB available storage (for AI models)

Download

Download Latest Release

1. Download SpeakType.dmg
2. Drag SpeakType to Applications
3. Grant Microphone + Accessibility + Documents Folder permissions
4. Download an AI model from Settings → AI Models

Press fn to start dictating.

Build from Source

git clone https://github.com/karansinghgit/speaktype.git
cd speaktype
make build && make run

---

Usage

1. Press hotkey (fn by default)
2. Speak your text
3. Release hotkey
4. Text is optionally enriched
5. Final text appears!

Tips:

• Speak naturally - Whisper handles accents well
• Say punctuation: "comma", "period", "question mark"
• Best results with 3-10 second clips
• Use Settings → AI to download or select your Ollama model before enabling local enrichment

---

Development

make build          # Build debug
make run            # Run app
make clean          # Clean build
make test           # Run tests
make dmg            # Create DMG installer

Current Issues

⚠️ When loading a model for the first time / switching to another model, there is a startup delay of 30-60 seconds.

So the first transcription will appear ultra slow, but it will go back to instantaneous dictation right after it's warmed up.

Project Structure

speaktype/
├── App/           # Entry point
├── Views/         # SwiftUI interface
├── Models/        # Data models
├── Services/      # Core functionality
├── Controllers/   # Window management
└── Resources/     # Assets & config

Tech Stack

• Swift 5.9+ / SwiftUI + AppKit
• WhisperKit - Local Whisper inference
• KeyboardShortcuts - Global hotkeys
• AVFoundation - Audio capture

---

Contributing

1. Fork & clone
2. Create a branch: git checkout -b feature/my-feature
3. Make changes and run make lint
4. Submit a PR

---

License

MIT License - see LICENSE for details.

---

Credits

• WhisperKit by Argmax
• OpenAI Whisper

---

Made with ❤️ for developers

*Privacy-first • Open Source *