TART: A Modular Tool for Technique-Aware Audio-to-Tablature Guitar Transcription

Standard automatic music transcription often discards the physical nuances of guitar performance, reducing expressive bends, slides, and fingering choices to plain pitch events. TART addresses this gap through a modular four-stage pipeline: audio-to-MIDI transcription, technique classification, audio-conditioned T5-based string-fret assignment, and automated tablature generation. Tested across GuitarSet, EGDB, and their noise-augmented counterparts, it achieves an average string-fret Tab F1 of 71.8% and an end-to-end Tab F1 of 54.08%, establishing the first direct path from raw audio to technique-annotated tablature.