Overview
music21 is a powerful Python-based framework designed for the analysis, transformation, and generation of music in symbolic formats such as MusicXML, MIDI, ABC, and Humdrum. Developed at MIT by Michael Cuthbert and a global community, it has evolved into the cornerstone of computational musicology. By 2026, music21 has solidified its position as the primary pre-processing engine for Symbolic Music Large Language Models (LLMs), allowing researchers to convert complex polyphonic scores into structured datasets suitable for tokenization and neural network training. Its technical architecture is centered around a hierarchical 'Stream' object model, which enables users to navigate through musical scores across multiple temporal and structural dimensions—such as parts, measures, chords, and notes—simultaneously. Unlike audio-processing libraries, music21 operates on the semantics of music, providing sophisticated tools for Roman numeral analysis, melodic contour detection, and rhythmic transformation. It integrates seamlessly with Jupyter Notebooks for interactive research and supports external engraving software like MuseScore and Finale for high-quality visualization of computed results. Its open-source nature ensures it remains the most extensible platform for academic research and professional music software development.