BEATs (Bidirectional Encoder representation from Audio Transformers) is a powerful audio classification model that achieves state-of-the-art results. This library provides a streamlined API for training and using BEATs models on audio classification tasks.
- 🤖 Automatic Checkpoint Management: Download BEATs models automatically
- 🎵 Feature Extraction: Extract high-quality audio embeddings
- 🔧 Simple Training API: Fine-tune BEATs with just a few lines of code
- 🏗️ Train from Scratch: Initialize and train BEATs without pre-trained weights
- � Long-Form Prediction: Process long audio files with timestamped predictions
- �📚 Comprehensive Documentation: Step-by-step guides and examples
- ⚡ GPU Support: Automatic CUDA detection and optimization
- Installation & Setup - Complete installation guide
- Feature Extraction - Extract audio embeddings
- Training Guide - Train models on custom datasets
- Long-Form Prediction - Process long audio files with timestamps
- Example Notebooks - Interactive tutorials
- Example Scripts - Ready-to-run code examples
MIT License. BEATs model weights are provided by Microsoft Research and OpenBEATs under their respective licenses.
Contributions welcome! See our development setup guide.
Need help? Check the documentation or open an issue.