macroroni 0.1.2

A speech-to-intent controller for a variety of VUI fields.

macroroni

macroroni is an open-source, highly adaptable speech-to-intent framework designed for Voice User Interface (VUI) development. It enables developers to seamlessly map spoken commands into actionable programmatic outputs or real-time game inputs (such as controls for classic emulation environments like OpenAI Gym/Gymnasium).

By combining local, cutting-edge Automatic Speech Recognition (ASR) engines with automated Machine Learning intent classification, macroroni bridges the gap between raw voice audio and immediate application state updates.

---

Features

• Multi-Backend Speech Recognition: Native support for high-performance localized transcription:
  • Apple Silicon (MPS): Optimized hardware acceleration using mlx-whisper.
  • NVIDIA GPU (CUDA): Fast, quantized execution using faster-whisper.
• Embedded Intent Classifier: Built-in NLP pipeline featuring a specialized TF-IDF Vectorizer coupled with an optimized, auto-calibrating Logistic Regression model.
• Low-Latency Architecture: Employs non-blocking, background thread audio listening combined with ambient noise microphone calibration to minimize latency.
• Extensible Architecture: Out-of-the-box support for modular customization (SingleVoiceController), including pre-built integrations for gaming wrappers like gym-super-mario-bros.

---

Requirements & System Support

macroroni is compatible with Python 3.8+ across major operating systems. Depending on your machine's hardware, ensure you install your preferred localized inference engine:

• For Apple Silicon Macs (M1/M2/M3/M4): pip install mlx-whisper
• For NVIDIA GPU Systems: pip install faster-whisper