Decomposer: Learning to Decompile Symbolic Music to Programs

2026-07-02

AI Summary

This work addresses the ill-posed inverse problem of generating executable and editable high-level music program instructions from symbolic music representations such as MIDI—a task hindered by the absence of paired data and a tendency to degenerate into mechanical transcription. The authors propose Decomposer, a novel framework that introduces the first synthetic paired corpus for Strudel-Synth and employs a two-stage strategy: initial mapping is learned via supervised fine-tuning, followed by joint optimization through reinforcement learning to balance MIDI reconstruction fidelity with the readability and diversity of the generated code. Experiments demonstrate that this approach significantly outperforms closed-source large language models in reconstruction quality on both synthetic and real-world MIDI benchmarks, while producing Strudel code that is more concise, diverse, and idiomatic than that generated by heuristic transpilers.

Abstract

Musical performance involves executing a set of high-level musical instructions, yet recovering those instructions from the performance is a challenging inverse problem. We present Decomposer, a post-training framework for symbolic music decompilation: the task of recovering executable, editable music programs from symbolic music. We instantiate the task as MIDI-to-Strudel decompilation, where the model takes symbolic MIDI as input and produces a program in Strudel, a music programming language, that reconstructs the input when executed. The task poses two challenges: Strudel is a low-resource language with little naturally paired MIDI-code data, and optimizing faithful reconstruction of MIDI alone can collapse to unreadable note-by-note transliteration. We address these challenges in two stages. First, we construct Strudel-Synth, a synthetic corpus of paired Strudel programs and rendered MIDI, and use it for supervised fine-tuning. Second, we refine the model with reinforcement learning on unpaired MIDI, optimizing rewards for both MIDI reconstruction faithfulness and code readability. Our evaluation across synthetic and real-world MIDI benchmarks shows that Decomposer achieves substantially higher MIDI reconstruction faithfulness than closed-source LLMs while producing more readable and diverse code than the heuristic converter.

Problem

• symbolic music decompilation
• MIDI-to-program recovery
• music program synthesis
• inverse problem in music

Innovation

• symbolic music decompilation
• Strudel programming language
• synthetic data generation
• reinforcement learning
• program synthesis