PaperlayerView exact revision on arXiv

Source-linked AI summary

How Far Can Chord-Symbol Time-Series Adaptation Carry Genre Identity? Capabilities and Boundaries in Multi-Genre Chord-Symbol Modeling

Jinju Lee

arXiv:2606.07334v4cs.SDcs.LG

TL;DR

This report asks whether one frozen chord model can be adapted to eleven genres through multiple parameter-efficient interfaces. Across the evaluation, adaptation improves held-out chord prediction, but no method is decisively superior and chord-symbol genre information remains incomplete.

  • Problem

    A practical composition tool needs one reusable frozen chord model that can be extended to new genres without separate full models.

  • Method

    The report evaluates five adaptation methods, control-token and wrong-genre diagnostics, classifiers, transition analyses, generated-output statistics, and real-song evaluation across eleven genres.

  • Results

    Every main method improves held-out target-genre chord prediction over the frozen pure-pop base, but no method is statistically dominant after correction.

  • Takeaways & Limitations

    Chord-symbol adaptation is useful but bounded: it reveals reusable genre-local information without establishing a decisive method winner or open-ended genre identity.

  • Takeaways & Limitations

    Automatic top-1 and top-5 prediction measure corpus-token agreement rather than musical quality, usability, or perceived genre authenticity, leaving listening-study validation missing.

Abstract

from arXiv · show

This revision updates an 11-genre chord-symbol adaptation report. The main 165-cell result is unchanged: all methods improve over the frozen pure-pop base, with no decisive method winner. v3 adds the ft-pop80-v2 multi-seed base-restoration note and corrects a few summary statistics for exact CSV faithfulness without changing conclusions.

Changes in v2

The revision clarifies that the released F1 base checkpoint is weight-identical to the pop-only Phase-0 baseline, while reported numbers remain unchanged. The fine-tuning run improved held-out jazz prediction, but checkpoint selection retained the pre-fine-tuning weights.

  • Revision scope: No reported number changes; the revision changes the released base artifact's description and the interpretation of the base-checkpoint ablation.
  • Checkpoint identity: 25,841,152 parameters were covered by an independent full SHA-256 comparison verifying the released F1 checkpoint's weight identity.The model contains 25,661,440 unique parameters because tied input/output embeddings are stored as two tensors.
  • Fine-tuning mechanism: Held-out jazz top-1 rose from 72.86 to 81.03 across epochs 3–8 during F1 fine-tuning.Best-checkpoint selection instead minimized validation loss on a pop-dominated mix, which rose from 0.5578 to 0.5901, retaining the Phase-0 initialization.
  • Interpretive consequences: All adaptation gains are gains over a pure-pop harmonic prior, and released evaluation numbers reproduce on independent CPU hardware within 0.02 pp.The base-checkpoint ablation is therefore a same-weights control whose ±0.4 pp spread estimates run-to-run noise.

Changes in v3

The revision adds a selection-corrected jazz-adapted base and corrects summary statistics against the released CSVs. These changes do not alter rankings, significance results, or qualitative conclusions.

  • Base restoration: Jazz top-1 reached 75.05 on the unified 9-source test for the hash-distinct selection-corrected ft-pop80-v2 base.
  • Multi-seed reproduction: 75.76 ± 0.03 jazz top-1 reproduced across three random seeds in matched-data retrains using the jazz-only validation selection rule.The base-robustness ablation is now limited by experimental effort rather than checkpoint availability.
  • Statistical corrections: Corrected Section 6.3 and Section 6.9 summary statistics restore exact faithfulness to the released CSVs.No ranking, significance result, or qualitative conclusion changes.

Changes in v4

The revision corrects LoRA's reported trainable footprint from a single rank-32 configuration to the rank range actually selected across genres. It also updates the associated multi-seed attribution.

  • LoRA footprint: LoRA uses 465,920 to 1,940,480 trainable parameters across ranks 4–64, corresponding to 1.8–7.6% of the model.The main grid selects rank per genre, rather than using only the previously listed rank-32 footprint.
  • Aggregate footprint: The reported overall footprint range changes from 0.9–4.5% to 0.9–7.6%.
  • Attribution correction: Sections 6.9 and 8 now attribute the three-seed 75.76 ± 0.03 reproduction to the selection-corrected jazz retrains.

1. Introduction

The report asks how much genre identity chord-symbol time series can carry and where that representation reaches its boundary. It evaluates adaptation methods as controllable probes of harmonic genre information, finding improvement without a decisive method winner.

  • 1. Introduction: The engineering goal is extending one frozen chord model to new genres without training and serving a separate full model for each genre.
  • 1. Introduction: The central question is how much genre identity chord-symbol time series can carry and where that representation reaches its boundary.
  • 1. Introduction: Chord symbols encode harmonic structure but omit rhythm, timbre, instrumentation, voicing, production, performance, lyrics, and listening context.The report therefore treats them as a deliberately limited controllable layer rather than a complete representation of genre.
  • 1. Introduction: LoRA, IA3, BitFit, prefix tuning, full fine-tuning, and control-token conditioning probe genre information available in chord-symbol sequences.
  • 1. Introduction: Every main method improves held-out target-genre chord prediction over the frozen base, but no method is statistically dominant after correction.The report frames this as a representation-boundary study rather than a method leaderboard.
  • 1. Introduction: The report contributes a complete 5-method × 11-genre × 3-seed evaluation plus diagnostics for method specificity, genre specificity, corpus size, and base choice.

2. Background

Chord symbols provide a compact, interpretable representation of harmonic motion, but genre identity also depends on rhythmic, timbral, vocal, and performance layers. The study therefore treats adaptation methods as probes of genre-local information in the chord-symbol layer rather than as a search for one best adapter.

  • Representation: Chord progressions encode key, time, bar, genre, and chord events while discarding voicing, fine-grained rhythm, timbre, melody, lyrics, and articulation.This abstraction supports shorter contexts, smaller vocabularies, and interpretable harmonic errors.
  • Representation: Top-1 and top-5 accuracy measure agreement with corpus continuations, not musical quality.The paper separates chord-prediction performance from perceived musical quality.
  • Representation: Genre identity may rely more strongly on groove, sound design, instrumentation, vocal style, or performance practice than on harmony, especially in hip-hop, electronic, funk, and pop-adjacent genres.
  • Research framing: Chord-symbol adaptation is expected to carry partial genre information: large genre-specific gains suggest usefulness, while saturation, comparable control-token performance, or low chord-only classification exposes representation boundaries.
  • Research framing: LoRA, IA3, BitFit, prefix tuning, and full fine-tuning are studied as adaptation interfaces, with control-token tuning as a lightweight baseline.The methods probe whether small interfaces can improve prediction over a frozen base and whether results reflect representation and data rather than one winning method.

3. Related Work

The report builds on symbolic chord modeling, style-conditioned progression generation, and lightweight controllable-generation methods. It distinguishes its contribution by using multi-genre adaptation and diagnostics to investigate what genre information remains in chord-symbol sequences.

  • Symbolic chord modeling: Prior chord-modeling research spans grammar-based, probabilistic, neural chord-generation, harmonization, Bach-chorale, and transformer-based systems.The Chordinator is identified as especially close because it studies style-conditioned chord progression generation across genres.
  • Positioning: This report extends prior style-conditioned work by comparing multiple adaptation methods across genres and mapping representation boundaries with diagnostics.The diagnostics include wrong-genre adapters, LoRA rank sweeps, and chord-focused evaluations.
  • Controllable generation: Recent controllable-generation studies examine rhythm and chord control, lightweight tuning, and chord-aware symbolic representations for multitrack generation.
  • Evaluation and data practice: Recent evaluation and dataset work motivates listener-centered assessment, human-preference alignment, de-duplication, and greater research openness.
  • Positioning: Unlike audio-domain LoRA work targeting timbral, rhythmic, and emotional control, this report isolates chord-symbol sequences to study retained genre information.Its base follows the Music Transformer family but uses a reduced chord-level vocabulary and sequence length.

4. Data and Representation

The study covers eleven heterogeneous target genres using normalized chord-symbol sequences and varied corpus sizes. Repetition and corpus construction constrain interpretation to held-out chord-transcription distributions rather than open-ended generalization.

  • Corpora and genres: The target set includes blues, bossa nova, Bach chorales, country, electronic, folk, funk, gospel, hip-hop, R&B/soul, and rock.Bach chorales are a separate tonal-chorale corpus and behave as a strong outlier in several results.
  • Base checkpoint: The released F1 base is weight-identical to the pop-only Phase-0 baseline, so the frozen base is effectively a pure-pop model.The prior jazz fine-tuning run improved held-out jazz top-1 from 72.86 to 81.03, but validation selection retained the pre-fine-tuning weights.
  • Corpora and genres: Corpus sizes range from 296 Bach-chorale sequences to 49,388 country sequences, motivating controlled-data-size comparisons.Bach chorales have 55 unique chords, while R&B/soul and funk have the highest chord entropy and lowest top-10 coverage.
  • Representation: The tokenizer extends a 351-token chord-symbol vocabulary to 359 tokens with additional genre markers and trainable genre embedding and output rows.
  • Representation: Chord notation is normalized into a canonical symbolic space before tokenization to support stable next-chord prediction and generation.
  • Data validity and repetition: Mean train-to-test 4-gram overlap is 0.975 across the eleven genres, while exact and key-normalized duplicate rates average 0.28% and 0.46%.Bach chorales reach 2.46% and 2.82% for those duplicate measures, respectively.
  • Data validity and repetition: The report limits its claim to held-out chord-transcription distributions and proposes novel-progression or progression-family splits as the next test of memorization sensitivity.If gains shrink on such splits, the boundary claim becomes sharper.

5. Method

The method compares five adaptation strategies and a control-token baseline across an 11-genre, three-seed evaluation grid. Parameter-efficient methods achieve performance close to full fine-tuning while using substantially fewer trainable parameters, with extensive diagnostics for reliability and representation boundaries.

  • Base and training setup: All main-grid experiments start from the same frozen 25.6M-parameter Music Transformer checkpoint, verified as identical to the pop-only Phase-0 baseline.
  • Base and training setup: LoRA, IA3, BitFit, prefix tuning, and full fine-tuning are trained for eight epochs across each target genre and three random seeds.LoRA ranks come from a prior sweep, and PEFT methods train the method-specific parameters plus new genre rows.
  • Method comparison: The four parameter-efficient methods land within roughly half a percentage point of full fine-tuning on macro top-1 while training only 0.9–7.6% as many parameters.Most methods match or exceed full fine-tuning, indicating little accuracy cost for modular adaptation.
  • Method comparison: Control-token tuning tests whether adapter gains can be accessed through a lightweight genre-conditioning interface over the frozen base.
  • Evaluation: The main metric is held-out next-token top-1 chord prediction, supported by top-5 and loss, across 11 genres, 5 methods, and 3 seeds.Method comparisons use corrected Wilcoxon signed-rank tests with Holm-Bonferroni and Benjamini-Hochberg correction.
  • Evaluation: Diagnostics cover LoRA reliability and rank selection, wrong-genre adapter rotation, chord-only classification, generated-output statistics, real-song chord-chart evaluation, and duplication.
  • Reproducibility: The complete grid and diagnostics were trained on a single RTX 4070 Laptop GPU with 8 GB memory, demonstrating feasibility without dedicated training infrastructure.
  • Reproducibility: Selected-rank LoRA results use highest validation top-1, whereas released adapters use lowest validation loss with a top-1 tiebreak, producing different ranks for six genres.Reproduction from released adapters should use the released-rank cells rather than the selected-rank row.

6. Results

Across the 165-cell grid, every method improves held-out top-1 over the frozen F1 base, but no method is decisively superior. The results support useful genre-local adaptation while showing that chord-symbol information and adaptation gains remain bounded.

  • 165 cells cover 5 methods × 11 genres × 3 seeds, and all five methods improve macro held-out top-1 over the frozen F1 base.
  • LoRA and IA3 lead macro top-1 and genre win count, but corrected pairwise tests do not support a decisive winner.
  • Control-token conditioning reaches macro top-1 82.01 and Delta +3.11 pp versus F1, placing it in the same accuracy band as adapter methods.
  • LoRA gains over F1 in all 11 genres, with a +3.71 pp macro gain, +2.74 pp median per-genre gain, and +2.52 pp non-chorale mean gain.
  • Rank changes usually have small effects outside Bach chorale, indicating that available genre information in chord-symbol sequences matters more than adapter rank alone.
  • Matched adapters exceed off-diagonal averages in 11/11 genres by +3.07 pp on average, while 81/110 off-diagonal cells still exceed the F1 baseline.
  • Adaptation moves generated unigram and bigram distributions closer to training data but lowers unique chord count and entropy in most genres, rather than demonstrating greater creativity.
  • Chord-only classification reaches accuracy 0.247 and macro F1 0.171 versus 11-class chance balanced accuracy 0.091, showing measurable but incomplete genre information.

7. Discussion

Chord-symbol adaptation provides useful, modular access to genre-local harmonic priors, but the representation captures only bounded genre information. The expanded experiments favor a representation-boundary interpretation over a LoRA leaderboard.

  • Small interfaces reliably improve held-out target-genre chord prediction across 11 genres over the frozen pure-pop base.The gains indicate that genre-local harmonic priors are present in the chord-symbol layer.
  • Adapters support modular deployment because genres can be added, replaced, or disabled without changing the full base model.The report identifies deployment, storage isolation, model versioning, and per-genre update cycles as additional adapter benefits.
  • Method differences are not decisive: control-token conditioning is strong, LoRA rank scaling is shallow in most non-chorale genres, and wrong-genre adapters frequently outperform F1.Generated LoRA outputs are closer to training distributions but often have lower entropy.
  • Chord symbols encode harmonic priors, common transitions, cadential habits, and chord-quality distributions, but omit groove, timbre, voicing, instrumentation, lyrical idiom, and production.These missing layers are described as likely essential for perceived genre authenticity.
  • When corpora are equalized, IA3 remains first, LoRA falls from second to last, and the four methods fall within about 0.9 pp.The report interprets the full-data LoRA edge as largely a data-availability effect rather than evidence of superior adaptation.
  • The report’s bounded conclusion is that chord-symbol adaptation reliably improves genre-local harmonic prediction, but chord symbols carry bounded rather than complete genre identity.

8. Limitations and Future Work

The study’s evidence is constrained by automatic evaluation, repetitive and coarse-grained data, omitted musical layers, and limited robustness testing. Future work should add perceptual, novel-progression, finer-grained, and broader baseline evaluations.

  • Automatic top-1 and top-5 prediction measure held-out token agreement, not musical quality, usability, or perceived genre authenticity.The real-song subset is not a listening study, leaving controlled listener or musician evaluation as the main missing validation.
  • Near-duplicate harmonic patterns are extremely common, limiting generalization claims beyond the corpus.The report recommends low-overlap or novel-progression evaluation as a future robustness check.
  • Genre labels are coarse, while Bach chorale is a tonal-chorale outlier rather than directly comparable to pop-adjacent genres.Chordonomicon-derived rock and country splits combine multiple substyles and transcription practices.
  • Chord symbols omit rhythm, voicing, texture, production, and timbre, which can dominate perceived genre identity.Future work should test perceived genre appropriateness using rendered progressions or musician inspection.
  • The Section 6.9 comparison is a same-weights control that calibrates run-to-run noise but provides no evidence about base-robustness.A genuine base ablation remains untested, although a jazz-adapted checkpoint now exists and its selection rule reproduced across three matched-data retrain seeds.
  • Reported accuracies depend on the flat root-quality tokenization, while adapter hyperparameters were only lightly swept.Different vocabularies or hyperparameter choices may shift the method-level numbers.
  • The report isolates the chord-symbol language-model contribution rather than downstream reranking, arrangement, or voice-leading layers.The deployed tool adds physics- and theory-retrieval modules and rule-based voicing outside the present scope.
  • Genre-conditioned statistical or retrieval baselines and public code, adapter, and evaluation-script releases would strengthen defensibility and reproducibility.The proposed comparisons include n-gram, Markov, and retrieval chord models.

9. Conclusion

The report studies chord-symbol adaptation as an interpretable middle layer for controllable music AI. Across eleven genres, adaptation improves chord prediction, but no method is decisively superior and chord-symbol genre information remains incomplete.

  • Multiple adaptation methods improve a frozen chord model across 11 target genres, with matched adapters showing consistent genre-local advantages.Because the released base is the pop-only Phase-0 baseline, every gain is over a pure-pop harmonic prior.
  • No method is decisively superior after correction, and equalizing corpus sizes reshuffles the method ranking.Control-token conditioning remains strong, chord-only genre classification weak, and generated-output diagnostics indicate distribution matching rather than open-ended diversity.
  • Chord-symbol sequences carry measurable harmonic genre information, enough to support modular adaptation, but not enough to replace rhythm, timbre, arrangement, or human perceptual evaluation.
  • Small conditioning over a shared harmonic base matters more robustly than the exact adapter family, while wrong-genre adapters often reveal a generic corpus-adaptation effect.
Loading 2606.07334v4…