MacParakeet addresses the need for a fast, private, and feature-rich voice interaction application on macOS. It provides system-wide dictation, file and YouTube transcription, meeting recording, and advanced text manipulation capabilities, all processed locally on Apple Silicon hardware. This makes it ideal for power users, developers, and privacy-conscious individuals seeking efficient voice-to-text solutions without relying on cloud services.

How It Works

MacParakeet leverages NVIDIA's Parakeet TDT model, optimized via FluidAudio CoreML to run on Apple's Neural Engine. For languages not covered by Parakeet, an optional local WhisperKit engine can be employed. The application orchestrates these speech-to-text (STT) engines through a shared runtime, ensuring all audio processing and transcription occur entirely on the user's Mac, preserving data privacy. Its architecture prioritizes speed, achieving approximately 155x real-time transcription, and efficient memory usage.

Quick Start & Requirements

• Primary Install: Download the notarized DMG from macparakeet.com and drag it to the Applications folder. Alternatively, install the CLI via Homebrew: brew install moona3k/tap/macparakeet-cli.
• Prerequisites: macOS 14.2+ and Apple Silicon (M1/M2/M3/M4).
• Setup: The first launch downloads the primary speech model (~6 GB) and speaker-detection assets (~130 MB). Subsequent operations are fully offline.
• Links: macparakeet.com

Highlighted Details

• System-wide dictation with hotkey support, push-to-talk, and optional media pausing.
• Comprehensive file and YouTube transcription with word-level timestamps and export to 7 formats (TXT, Markdown, SRT, VTT, DOCX, PDF, JSON).
• Meeting recording capabilities, including system audio, microphone input, live transcript preview, and calendar integration for reminders and auto-start.
• "Transforms" feature for on-demand text rewriting using configurable local or cloud-based LLM providers (OpenAI, Anthropic, Gemini, Ollama, etc.).
• Performance metrics include ~155x realtime transcription, ~2.5% word error rate (Parakeet TDT 0.6B-v3), and ~66 MB working memory per active inference slot.
• Supports 25 European languages via Parakeet, with additional multilingual support through WhisperKit.

Maintenance & Community

The project encourages contributions via GitHub issues for bug reports and feature discussions before submitting pull requests. Sponsorship is welcomed as a means of support. Specific community channels like Discord or Slack are not mentioned in the README.

Licensing & Compatibility

The project is licensed under GPL-3.0. This free software license imposes copyleft restrictions, meaning derivative works must also be made available under the GPL-3.0 license. Commercial use or integration into closed-source applications may require careful consideration of these terms.

Limitations & Caveats

MacParakeet is exclusively compatible with Apple Silicon Macs (M1/M2/M3/M4). The default Parakeet TDT engine is optimized for English and European languages; support for other languages necessitates downloading separate WhisperKit models. Initial setup requires downloading substantial model files.