symbolic music representation

This study investigates whether large language models (LLMs), trained exclusively on text without explicit musical supervision, can implicitly acquire representations of symbolic music structure. Methodologically, we prompt LLMs with plain-text instructions to generate MIDI sequences, construct a symbolic music dataset from these outputs, and train dedicated neural networks for downstream tasks—including music genre classification, style identification, and melody completion. We provide the first empirical evidence that LLMs develop internal representations of core musical concepts—such as pitch, rhythm, and tonality—as demonstrated by the viability of their generated MIDI data for supervised learning: classification accuracy approaches that of fully supervised baselines, and melody completions exhibit musically coherent phrasing and harmonic consistency. Results indicate that LLMs possess nascent—but incomplete—reasoning capabilities over symbolic music structure, establishing a novel paradigm for zero-shot and few-shot music generation and understanding.

This paper proposes a label-free, unified modeling paradigm for symbolic music, framing diverse tasks—including chord recognition, melody generation, and drum pattern generation—as a single “music-to-music” sequence modeling problem. Methodologically, it introduces (1) a novel *function alignment* paradigm that implicitly encodes functional semantics (e.g., harmonic role, rhythmic function) via intrinsic structural patterns in note sequences, eliminating reliance on external annotations; (2) a lightweight, learnable adapter that bridges two pre-trained language model encoders, enabling parameter-efficient, task-agnostic modeling; and (3) a symbolic-music-specific tokenization scheme integrated with functional semantic alignment. The framework achieves state-of-the-art performance across multiple symbolic music understanding and generation benchmarks. All code, pretrained models, and interactive demos are publicly released.

Unsupervised discovery of “music words”—reusable, structurally coherent musical fragments—in symbolic music remains challenging due to inherent semantic ambiguity. Method: We propose a sparse coding–based statistical modeling framework that formalizes music word discovery as an optimization problem minimizing total encoding length, thereby aligning with human perceptual coding principles and mitigating ambiguity. Our approach employs a two-stage EM algorithm integrating sparse representation learning with pattern dictionary learning, enabling interpretable decomposition and reconstruction of musical sequences. Contribution/Results: Evaluated on a multi-style dataset, the discovered music words achieve an IoU of 0.61 against expert annotations and demonstrate strong cross-style generalization. Beyond providing interpretable, structure-aware priors for music generation, classification, and style transfer, this work establishes the first information-compression–driven paradigm for structural analysis in computational musicology.

Existing music generation systems lack structured scrutiny of audio outputs and controllable editing capabilities. This work proposes Libretto, a novel framework that, for the first time, translates symbolic music into a structured, editable, and quantifiable representation amenable to language models. By introducing a symbolic syntax explicitly modeling note onsets, voice assignments, and measure structures tailored for large language models (LLMs), Libretto establishes a multidimensional statistical evaluation space encompassing rhythm, harmony, and melody. The framework enables multitask generation, retrieval-based diagnostics, copyright risk mitigation, and iterative self-refinement along musically meaningful structural axes. It significantly enhances the structural coherence and controllability of generated outputs across diverse tasks, including inpainting, reference-guided composition, progressive transformation, and educational music generation.

This work addresses the limitations of existing symbolic music representations, where grid-based encodings suffer from low efficiency and data sparsity, while pure event sequences struggle to preserve structural invariance and spatial locality. To overcome these issues, the authors propose a novel hybrid representation that integrates piano-roll and event-sequence paradigms through four complementary event types: Frame, Gap, Pattern, and Musical Structure. This design effectively captures temporal dependencies, local spatial features, and higher-level structural information. The resulting representation achieves a superior trade-off between sequence length and vocabulary size, yielding a 1.36–7.16× improvement in encoding efficiency. Experiments using autoregressive generative models demonstrate that the proposed method significantly outperforms state-of-the-art baselines in both automatic metrics and human evaluations.

This study addresses the lack of unified, reproducible information-theoretic tools in symbolic music analysis, particularly for harmonic progressions and rubato modeling. We present vega-mir, an open-source Python toolkit integrating nine quantitative measures—including Shannon entropy, Kullback–Leibler divergence, Zipf’s law fitting, chord transition graph networks, and rubato spectral analysis—to enable systematic large-scale corpus studies. Innovatively combining PageRank with KL divergence, we reveal for the first time a significant correlation (ρ = 0.61) between harmonic centrality and stylistic deviation across 14 MAESTRO composers. Furthermore, applying Higuchi fractal dimension and periodicity spectrum analysis, we challenge the stereotype of Glenn Gould’s “mechanical” timing, demonstrating that his performances exhibit the strongest rubato periodicity among three pianists, with a median dominant period of 66 beats.

This work addresses the limited controllability of existing symbolic music generation methods, which typically rely on from-scratch synthesis and struggle to support explicit local editing. The paper introduces the first explicit editing framework for symbolic music, reframing generation as a draft-editing process. Built upon a BEAT-based rhythmic grid anchoring scheme, the framework unifies three editing mechanisms—token-wise sequence labeling, iterative accompaniment refinement, and post-hoc token infilling—within a single pre-trained backbone model to enable efficient inference. Experimental results demonstrate that the proposed approach outperforms both autoregressive and diffusion models across three distinct editing tasks, achieving inference latency under 100 milliseconds while significantly improving both generation accuracy and perceptual audio quality. These findings highlight a strong correlation between music representation design and editing efficacy.

This study addresses the challenge that existing deep music generation models struggle to perceive the distinct roles of individual voices in polyphonic music. Focusing on Haydn’s “Lark” Quartet, the work integrates auditory analysis, electroacoustic measurement, and machine-based representational reconstruction to establish a mapping from classical contrapuntal structures to computable AI features. It introduces a novel “role-aware encoding” mechanism that abandons conventional quantized grids in favor of event-based timestamps to capture micro-temporal nuances, while transforming acoustic features into phenomenological anchors informed by aesthetic intuition. By combining DAW-based spectral and dynamic analysis, event modeling, and symbolic generation techniques, the research achieves a coherent loop linking classical music analysis, electronic representation, and AI-driven composition, thereby laying a theoretical foundation for human–AI collaborative systems endowed with social awareness and intersubjective musical understanding.

Current evaluations of symbolic music in large language models suffer from a lack of standardized representations, datasets, and metrics. To address this gap, this work proposes LilyBench—the first comprehensive benchmark based on LilyPond—which unifies music generation and understanding tasks within a single symbolic framework and introduces a multidimensional evaluation strategy. Integrating LilyPond notation, MusPy descriptors, Jensen-Shannon similarity, LilyBERT embeddings, and Fréchet Music Distance, the benchmark enables compilable score generation under zero-shot settings. Experimental results demonstrate that existing open-source large models excel at composer and genre identification but still face significant challenges in structural music understanding. The study also reveals systematic discrepancies among different evaluation metrics, highlighting the need for more coherent assessment methodologies in symbolic music modeling.

Existing symbolic music alignment formats support only one-to-one mappings, making them inadequate for representing complex many-to-many alignment scenarios such as repeated rehearsal passages or figured bass improvisation. This work proposes a lightweight, backward-compatible extension to the Match format that enables flexible encoding of many-to-many alignments within a standard framework by introducing virtual score tokens and pointers to performed notes. Additionally, the “section” directive is enhanced to annotate semantically meaningful segments that transcend explicit score indications. The approach is validated on two representative use cases—piano rehearsal and figured bass realization—demonstrating its capacity to balance expressive power with compatibility in existing alignment systems.