AutoTab: An Interactive System for Automatic Guitar Tablature Transcription

This work introduces AutoTab, an interactive system designed to simplify the process of transcribing music into guitar tablature. By combining sound event detection, Convolutional Neural Networks (CNNs), and graph-based optimization algorithms, AutoTab offers an intuitive and accessible solution for musicians of all skill levels. The system employs precise note and chord identification through CNNs, a WebSocket-based server for real-time audio analysis, and specialized algorithms such as the Audio Window Analyser for improved sound event detection and TabGen for optimized tablature generation. Initial results demonstrate robust performance, with the Notes Model achieving 87.8% accuracy in note identification and the Chords Model achieving a 78% Macro F1-score in chord recognition. AutoTab aims to democratize access to high-quality transcription technology, enhancing musical practice and learning for guitarists worldwide. Future work will focus on addressing the challenges in polyphonic transcription and expanding the system’s functionality.