Pith. sign in

REVIEW 4 major objections 5 minor 57 references

MAJEPPA: Morphing and Assessing in a Unified Piano Performance Space

T0 review · 4 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read MAJEPPA claims a single score-conditioned model can generate and evaluate piano performances across the full skill spectrum.

desk verdict MAJEPPA brings a genuinely useful piano dataset and a credible JEPA extension, but the claimed wins over baselines hinge on significance tests and a train/test overlap control that are missing. read the letter →

arxiv 2608.11026 v1 pith:YPHM44FA submitted 2026-08-11 eess.AS cs.MM

Jinwen Zhou , Huan Zhang , Weixi Zhai , Jinhua Liang , Aidan O. T. Hogg , Simon Dixon This is my paper · ORCID
classification eess.AScs.MM
keywords pianoperformancerepresentationlearningJEPAautoregressivetransformerscore-performancealignmentassessmentmistakepredictionsymbolicmusic
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

MAJEPPA aims to show that one score-conditioned model can span the entire piano-performance space, from a child's first practice attempt to a virtuoso's concert recording, by learning a representation that supports both generation and assessment. To do this, the authors curate roughly 4,000 annotated YouTube recordings across six expertise levels and six recording contexts, transcribe them to MIDI, align each performance segment to its score, and train a pre-trained autoregressive MIDI transformer with a joint generative-plus-JEPA objective. The paper introduces EVPMR, a four-task benchmark for piano-performance understanding, and reports that frozen MAJEPPA embeddings beat generative, masked-reconstruction, and contrastive baselines on Chopin competition ranking, technique classification, quality regression, and conspicuous-mistake prediction. If the claim holds, music-education systems could use a single model to demonstrate how a piece should sound, diagnose where a student's playing goes wrong, and track progress over time.

What carries the argument

The load-bearing mechanism is a two-view Joint-Embedding Predictive Architecture (JEPA): instead of reconstructing the performance, the model predicts a latent performance embedding from the score alone, while a parallel generative loss keeps the model able to produce performance tokens. A score–performance pair $(s,p)$ is packed into one token sequence and passed through the same transformer twice. Pass 1 uses a causal mask and computes next-token cross-entropy over the performance tokens, preserving score-conditioned generation. Pass 2 uses a block-diagonal causal mask that makes the score view $[s; \text{COND}; \text{PRED}]$ and the performance view $[p; \text{EOS}]$ mutually invisible, forcing the [PRED] token to produce $\text{Pred}(s)$ without observing the performance. A shared linear projection maps both $\text{Pred}(s)$ and $\text{Enc}(p)$ into an $\ell^2$-normalized 512-dimensional space, where three contrastive terms act: InfoNCE piece alignment $L_{\text{piece}}$, supervised cross-condition clustering $L_{\text{cond}}$ with ordinal-distance weights, and symmetric within-performance InfoNCE $L_{\text{self}}$. The generative loss prevents representation collapse, and the contrastive losses give the embedding space musical structure.

What would settle it

Search for exact or near-duplicate MIDI or audio matches between the 3,979 MAJEPPA recordings and the EVPMR evaluation recordings (PISA, NeuroPiano, YCU-PPE-III, Chopin, Pianism, Pianist8); if any overlap exists, rerun the linear-probe benchmark with those clips removed and compare MAJEPPA's margins against the Aria and Moonbeam baselines.

Watch

Extended reading notes

Core claim

The paper claims that a single autoregressive transformer, pre-trained on symbolic MIDI and adapted with LoRA, can learn a unified piano-performance space by solving two tasks at once: score-conditioned next-token prediction over performance tokens, and latent prediction of a performance embedding from its score alone. The latent prediction uses a JEPA-style two-view setup: score and performance are encoded under a block-diagonal causal mask so the [PRED] token never sees performance tokens, and three contrastive losses pull predicted score embeddings toward their paired performance embeddings (piece alignment), cluster performances by expertise and recording context (cross-condition clustering), and keep segments of the same performance coherent (within-performance alignment). On the new EVPMR benchmark, the frozen embeddings beat four baselines on Chopin competition ranking, technique classification, quality regression across PISA, NeuroPiano, and YCU-PPE-III, and mistake prediction on two of three datasets. The paper therefore asserts that generation and understanding are not competing modes but two outputs of one score-conditioned representation.

Load-bearing premise

The benchmark numbers assume that no MAJEPPA pre-training clip, especially one sourced from YouTube, also appears in the EVPMR test sets, since the paper never states that overlapping recordings were removed.

Editorial extensions

If this is right

  • If the central claim is right, a single frozen embedding can rank Chopin competition performances, detect piano techniques, score quality on three datasets, and predict conspicuous mistakes without task-specific fine-tuning.
  • The same model that produces these embeddings can also generate score-conditioned performances, so an education tool could both demonstrate how a piece sounds at different skill levels and diagnose where a student's playing diverges from the score.
  • The MAJEPPA dataset fills the beginner-to-virtuoso gap, enabling models that have actually seen practice, sight-reading, and teacher demos rather than only concert recordings.
  • The ablation suggests the three contrastive losses have a division of labour, so future systems can tune representation objectives per downstream task.
  • Because the best full model improves over the Aria and Moonbeam baselines, the gain comes from the score-conditioned JEPA objective, not from model scale or tokenizer design.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Beyond the paper, the same two-view score–performance setup could be applied to other instruments or to audio-only recordings by treating a cover or a different rendition as the second view, since it needs only paired renderings of the same musical content.
  • Beyond the paper, the apparent monotonic drift in practice trajectories hints that MAJEPPA embeddings could serve as a continuous progress metric for learners, not just a categorical six-level label; a direct test would correlate embedding distance with external exam grades.
  • Beyond the paper, the method suggests a controllable 'morphing' interface—interpolating between a beginner and a virtuoso embedding to generate intermediate skill levels—but the paper does not implement or evaluate such interpolation.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper introduces MAJEPPA, a dataset of 3,979 YouTube piano recordings manually annotated along two axes (performer expertise and recording context) and aligned to scores via subsequence DTW. The proposed method adapts Aria-medium through LoRA with a joint objective: a causal next-token generation loss and three contrastive JEPA terms (L_piece, L_cond, L_self) computed under a block-diagonal attention mask with [PRED] tokens. The authors also introduce the EVPMR benchmark and evaluate frozen embeddings with linear probes on competition ranking, technique classification, quality regression, mistake prediction, and pianist identification. The central claims are that MAJEPPA variants outperform four baselines across EVPMR tasks and that the same model both generates and understands piano performance.

Significance. The dataset is a potentially valuable resource: it is, to my knowledge, the first large-scale piano-performance corpus spanning child beginners to virtuosi, and the EVPMR benchmark offers a reusable evaluation suite. The method is clearly described, the two-pass attention masking is an interesting way to combine generation and representation learning, and the three-loss ablation is a useful analysis. The paper is also candid about its main limitations. However, the headline empirical claims are not yet supported: no train/test disjointness control is reported, no significance testing accompanies Table 2, and the generation side of the 'unified' claim is explicitly unevaluated. These are fixable with additional controls, which is why I recommend major revision rather than rejection.

major comments (4)
  1. [Section 3 and Section 5] The manuscript does not state that the MAJEPPA training corpus and the EVPMR test sets are disjoint. Section 3 collects YouTube recordings by querying piece titles with keywords such as 'Practice', 'Sight-reading', and 'Progress' and includes concert recordings of ATEPP repertoire; Section 5 evaluates on PISA (59 YouTube videos), Pianist8, and the 2015 Chopin competition data. A YouTube practice video in PISA, or a concert recording used for Pianist8 or the Chopin set, could also be among the 3,979 training recordings. Because MAJEPPA is LoRA-finetuned on those recordings, any overlap lets the frozen embeddings memorize identity markers and inflate linear-probe accuracy. Please add an explicit deduplication procedure (e.g., video-ID or audio-hash matching) and either exclude overlapping recordings from training or report results on a verified-disjoint split.
  2. [Table 2 and Section 5] The claim that 'MAJEPPA variants outperform various baselines' is not supported by the reported statistics. The table gives mean (std) over 5-fold cross-validation across 5 seeds but no significance tests. Several improvements are comparable to the reported standard deviations, for example Technique Acc 79.0 (6.5) versus Moonbeam 77.9 (4.8) and Pianist8 86.8 (5.2) versus Aria 86.1 (4.3); on Pianist8 the full model is below Aria. Because the 25 evaluation runs are paired by fold and seed, a corrected paired t-test or a non-parametric equivalent should be reported for all task-model comparisons, together with effect sizes or confidence intervals.
  3. [Section 5.1] The abstract and introduction state that MAJEPPA 'both generates and understands performances in a unified framework,' but the paper explicitly says 'we did not conduct a full-scale evaluation of the generation capabilities' and only points to a website. Since the generative loss is a core component and the title claims morphing as well as assessing, quantitative generation results (objective metrics or a human listening test against the Aria or Moonbeam baselines) are needed before the unified-generation claim can be evaluated. If generation is out of scope, the claims should be narrowed to representation learning.
  4. [Section 3.2] The parenthetical 'we did not explicitly check transcription quality (including pedal artefacts)' is a load-bearing limitation because all audio-only datasets in Section 5 are transcribed with Transkun before computing every downstream metric, and the MAJEPPA pre-training pairs are also built from Transkun transcriptions. If AMT errors are systematic, for instance at low-skill recordings with more overlapping notes or pedal blur, they could confound the reported quality-regression and mistake-detection improvements. Please validate transcription on a small annotated subset, report error rates by expertise and context, and state whether the main conclusions in Table 2 are robust to excluding the noisiest transcriptions.
minor comments (5)
  1. [Section 4.3] The definition of w_ij in Eq. (3) is described only verbally and the exact closed form is deferred to a website; for reproducibility, include the formula in the paper.
  2. [Section 3.1] The annotation procedure references a calibration document and a reliability website but reports no inter-annotator agreement; report Cohen's kappa or Krippendorff's alpha for the expertise and context labels, since L_cond is supervised by these labels.
  3. [Section 5.1] The practice-progression analysis reports a mean Spearman correlation of rho = 0.41 across PC1 and PC2 for only 7 YouTube journeys; report per-journey correlations and a shuffle baseline to support the claim of directional monotonic drift.
  4. [Table 2] The UMP columns are compared only against Aria because other baselines use incomparable token schemes; this should be stated in the table caption or as a column note so the reader does not infer a full baseline comparison.
  5. [Abstract and Section 4.3] The term 'self-supervised' is used loosely: L_cond is a supervised contrastive loss using annotator-provided expertise and recording-context labels, so the overall framework is better described as self-supervised with light annotation or semi-supervised.

Circularity Check

0 steps flagged · score 0.0 of 10

Central claim is independent of training objectives; no circularity by construction, though evaluation-corpus overlap is an unquantified validity risk.

full rationale

No circularity found. The central pipeline is: MAJEPPA pairs (score, performance, condition tags) are used to fine-tune Aria with a next-token objective plus three contrastive terms; downstream EVPMR probes are applied to frozen encode() embeddings from independent or prior-published datasets (NeuroPiano, YCU-PPE-III, PISA, PianoJudge's Chopin/technique sets, Pianist8, UMP corpora). None of the EVPMR labels is also a training loss input: quality ratings, competition rounds, technique classes, pianist identity, and mistake labels are not used in Eq. (1). The supervised contrastive term Lcond is trained on MAJEPPA's coarse expertise/context annotations, which correlate with skill but are not the downstream rating values; this is transfer learning, not derivation-by-construction. The paper's many self-citations (ATEPP, NeuroPiano, PianoJudge, Moonbeam, AMT-JEPA) serve as data repositories, baselines, or related work; no uniqueness claim is imported from them and nothing load-bearing reduces to a self-citation. The one legitimate validity risk—the evaluation sections never state that MAJEPPA training recordings were removed from the YouTube-sourced PISA or other test sets—would be data leakage rather than circularity by construction, and it is not established by the paper's equations. Under the hard rule that circularity must be exhibited as a specific reduction, this paper does not warrant a circularity score above 0.

Assumptions & free parameters 8 free parameters · 5 assumptions · 0 invented entities

The central method is a machine-learning pipeline rather than a mathematical derivation, so the ledger records human-chosen loss weights and unchecked domain assumptions that the benchmark results rely on. No new physical or conceptual entities are introduced.

free parameters (8)
  • gamma (L_gen weight) = 1.0
    Balance between generative loss and contrastive JEPA terms; chosen by ablation, reported in Section 4.3.
  • lambda (L_piece weight) = 8.0
    Weight of the score-performance InfoNCE alignment loss; chosen by ablation; strongly shapes the representation.
  • beta (L_cond weight) = 1.0
    Weight of the cross-condition supervised contrastive loss; chosen by ablation.
  • alpha (L_self weight) = 1.0
    Weight of within-performance segment alignment; chosen by ablation.
  • temperature tau = 0.07
    Temperature for all contrastive losses; standard value chosen by ablation.
  • number of PRED tokens k = 1
    Number of query tokens for the JEPA predictor; chosen by ablation.
  • LoRA rank and alpha = r=512, alpha_LoRA=32
    Model adaptation capacity; chosen by the authors without ablation details.
  • learning rates and epochs = 2e-4, 1e-3, 1e-5; 120 epochs
    Optimization hyperparameters that affect final embeddings; reported but not varied in the paper.
assumptions (5)
  • domain assumption Aria-medium is a suitable pretrained backbone and tokenizer for noisy, real-world piano MIDI.
    The method freezes Aria weights and only trains LoRA adapters, so any mismatch between Aria's pretraining distribution and Transkun-transcribed YouTube audio would propagate through the representation.
  • domain assumption Transkun transcription is accurate enough for both pre-training and evaluation without manual correction.
    Section 3.2 explicitly states transcription quality was not checked, yet MIDI from Transkun is used as the target view for the predictive losses and as the representation for all EVPMR tasks.
  • domain assumption The manual expertise and recording-context annotations are consistent.
    Section 3.1 describes a shared calibration document and reference clips but reports no inter-annotator agreement statistics; these labels drive the supervised contrastive loss Lcond.
  • domain assumption Subsequence DTW on chroma features reliably aligns score and performance despite mistakes and omitted repeats.
    Section 3.2 uses this alignment to build score-performance pairs; the authors acknowledge that extra repetitions can produce locally noisy pairs.
  • domain assumption A deadpan score tokenization and two conditioning tags capture the main variation in piano performance.
    The two-view JEPA formulation assumes that score plus performer and context tags is a sufficient context to predict the abstract performance embedding; unmodeled factors such as pedaling and microphone quality are folded into the target.

how reviews work

0 comments
Cite this review

Pith. "Pith review of MAJEPPA: Morphing and Assessing in a Unified Piano Performance Space." pith.science (2026). https://pith.science/paper/YPHM44FA

@misc{pith2026260811026,
  author       = {Pith},
  title        = {Pith review of: MAJEPPA: Morphing and Assessing in a Unified Piano Performance Space},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YPHM44FA}},
  note         = {Machine review of arXiv:2608.11026}
}
read the original abstract

We present MAJEPPA, a self-supervised framework to learn piano performance representations that span the full skill spectrum, from beginner practice sessions to virtuoso concert recordings. We curate the MAJEPPA dataset, comprising ~4,000 annotated recordings across six expertise levels and six recording contexts. We adapt a single pre-trained MIDI autoregressive model with a joint objective: next-token prediction learns score-conditioned performance generation at various skill levels, while InfoNCE and supervised contrastive losses align abstract score and performance representations in a joint embedding space. The proposed model both generates and understands performances in a unified framework. By introducing the EVPMR benchmark, a suite of downstream tasks spanning quality assessment, competition ranking, mistake and technique classification, we evaluate the learnt representations, demonstrating progress towards a real-world model for the piano performance space.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

57 extracted references · 52 canonical work pages

  1. [1]

    MAJEPPA: Morphing and Assessing in a Unified Piano Performance Space

    INTRODUCTION Piano performance is a remarkably rich domain of human expression. The same score can be realised as a child’s halt- ing first attempt, an adult student sight-reading, a teacher demonstrating phrasing at half tempo, or a virtuoso com- manding the concert stage. Understanding this variation computationally, distinguishing not justwhatis played...

    work page 2026

  2. [2]

    TheMAJEPPA dataset, the first large-scale dataset to span the full performance spectrum from child beginners to virtuosi

    work page

  3. [3]

    The firstJEPAto model for joint symbolic piano- performance generation and representation learning, using score–performance pairs as a natural two-view structure

    work page

  4. [4]

    TheEVPMRbenchmark suite, designed for evalua- arXiv:2608.11026v1 [eess.AS] 11 Aug 2026 tion across downstream tasks, including Chopin com- petition ranking, technique detection, performance quality regression, and conspicuous mistake predic- tion, where our models outperform various baselines

    work page Pith review arXiv 2026

  5. [5]

    RELA TED WORK 2.1 Music and MIDI representation learning Approaches to learning representations from symbolic mu- sic are shaped by how MIDI is serialised into a sequence, and tokenisation schemes such as REMI [17], Compound Word [18], OctupleMIDI [19], and PerTok [20] provide the input vocabularies on top of which the following families of self-supervise...

    work page

  6. [6]

    DA TA CURA TION Existing symbolic music datasets focus predominantly on professional performances: ATEPP [13] contains virtuoso interpretations and ASAP [14] provides competition-level recordings with aligned scores. To address this gap, we cu- rate theMAJEPPA dataset, comprising 3,979 real-world piano recordings across 942 distinct pieces and movements, ...

    work page

  7. [7]

    PROPOSED METHOD Our goal is to learn a unified representation of piano per- formance that captures both the underlying musical content and its expressive realisation. To this end, we model score and performance as two complementary views of the same musical excerpt, and learn to predict performance represen- tations from score representations in a shared ...

    work page

  8. [8]

    conspicuous mistake

    DOWNSTREAM TASKS AND EV ALUA TION Inspired by EV AR, we formulate performance-related tasks in a unified package, EVPMR, providing standardised eval- uation protocols. All audio-only datasets are transcribed to MIDI with Transkun [39] for consistency. Every task is evaluated with 5-fold cross-validation repeated across 5 random seeds. On top of a frozen g...

    work page 2015

Show all 57 references
  1. [9]

    We contribute ∼4,000 annotated recordings across six expertise levels and six recording contexts, spanning from child beginners to concert virtuosi

    CONCLUSION We presented MAJEPPA, a dataset and framework for uni- fied piano performance representations. We contribute ∼4,000 annotated recordings across six expertise levels and six recording contexts, spanning from child beginners to concert virtuosi. We use the dataset to ...

  2. [10]

    The corpus includes children’s practice and showcase videos and commercial or concert recordings by professional pianists

    ETHICS STA TEMENT MAJEPPA is built from publicly available YouTube solo- piano recordings for research purposes; because these recordings are publicly accessible, we did not obtain individ- ual consent from uploaders. The corpus includes children’s practice and showcase videos...

  3. [11]

    ACKNOWLEDGMENTS This work is supported by the UKRI Centre for Doctoral Training in Artificial Intelligence and Music. J. Zhou is a research student supported jointly by the China Scholarship Council and Queen Mary University of London

  4. [12]

    Sounds out of pläce? Score-Independent de- tection of conspicuous mistakes in piano performances,

    A. Morsi, K. Tatsumi, A. Maezawa, T. Fujishima, and X. Serra, “Sounds out of pläce? Score-Independent de- tection of conspicuous mistakes in piano performances,” inProceedings of the International Society for Music Information Retrieval Conference (ISMIR), 2023, pp. 352–358

    2023

  5. [13]

    Simulating piano performance mistakes for music learning,

    A. Morsi, H. Zhang, A. Maezawa, S. Dixon, and X. Serra, “Simulating piano performance mistakes for music learning,” inProceedings of the 21st Sound and Music Computing Conference (SMC), 2024, pp. 179– 186

    2024

  6. [14]

    RUMAA: Repeat-aware unified music audio analysis for score- performance alignment, transcription, and mistake de- tection,

    S. Chang, S. Dixon, and E. Benetos, “RUMAA: Repeat-aware unified music audio analysis for score- performance alignment, transcription, and mistake de- tection,” in2025 IEEE Workshop on Applications of Signal Processing to Audio and Acoustics (WASPAA), 2025, pp. 1–5

    2025

  7. [15]

    How does the teacher rate? Observations from the NeuroPiano dataset,

    H. Zhang, V . Cheung, H. Nishioka, S. Dixon, and S. Fu- ruya, “How does the teacher rate? Observations from the NeuroPiano dataset,” inInternational Society for Music Information Retrieval (ISMIR) Late Breaking Demo (LBD), 2024

    2024

  8. [16]

    From audio en- coders to piano judges: Benchmarking performance understanding for solo piano,

    H. Zhang, J. Liang, and S. Dixon, “From audio en- coders to piano judges: Benchmarking performance understanding for solo piano,” inProceedings of the 25th International Society for Music Information Re- trieval Conference (ISMIR), 2024, pp. 511–519

    2024

  9. [17]

    LLaQo: Towards a query-based coach in expressive music performance assessment,

    H. Zhang, V . K. Cheung, H. Nishioka, S. Dixon, and S. Furuya, “LLaQo: Towards a query-based coach in expressive music performance assessment,” inICASSP 2025 - 2025 IEEE International Conference on Acous- tics, Speech and Signal Processing (ICASSP), 2025, pp. 1–5

    2025

  10. [18]

    Modeling the difficulty of sax- ophone music,

    Š. Libˇrický and J. Hajiˇc, “Modeling the difficulty of sax- ophone music,” inProceedings of the 26th International Society for Music Information Retrieval Conference (IS- MIR), 2025, pp. 747–754

    2025

  11. [19]

    Playability prediction in digital guitar learn- ing using interpretable student and song representations,

    M. Müllerschon, A. Klapuri, M. Rodriguez, and C. Cardin, “Playability prediction in digital guitar learn- ing using interpretable student and song representations,” inProceedings of the 26th International Society for Mu- sic Information Retrieval Conference (ISMIR), 2025, pp. 631–637

    2025

  12. [20]

    What song now? Personalized rhythm guitar learning in western popular music,

    Z. Hassein-Bey, Y . Abbou, A. D’Hooge, M. Giraud, G. Guillemain, and A. Jeanneau, “What song now? Personalized rhythm guitar learning in western popular music,” inProceedings of the 26th International Society for Music Information Retrieval Conference (ISMIR), 2025, pp. 296–302

    2025

  13. [21]

    Enabling empirical analysis of piano perfor- mance rehearsal with the Rach3 MIDI dataset,

    A. Morsi, S. Chiruthapudi, S. D. Peter, M. Pilkov, L. Bishop, A. Maezawa, X. Serra, and C. E. Cancino- Chacón, “Enabling empirical analysis of piano perfor- mance rehearsal with the Rach3 MIDI dataset,” inPro- ceedings of the 26th International Society for Music Information Re...

    2025

  14. [22]

    PianoV AM: A multimodal piano perfor- mance dataset,

    Y . Kim, J. Park, J. Bae, K. Kim, T. Kwon, A. Lerch, and J. Nam, “PianoV AM: A multimodal piano perfor- mance dataset,” inProceedings of the 26th International Society for Music Information Retrieval Conference (IS- MIR), 2025, pp. 528–535

    2025

  15. [23]

    Predicting flutist onset timing in duet performance: A multimodal analysis of gesture and breath cues,

    J. Choi, T. Kwon, and J. Nam, “Predicting flutist onset timing in duet performance: A multimodal analysis of gesture and breath cues,” inProceedings of the 26th International Society for Music Information Retrieval Conference (ISMIR), 2025, pp. 100–106

    2025

  16. [24]

    ATEPP: A dataset of automatically transcribed expressive piano performance,

    H. Zhang, J. Tang, S. R. Rafee, S. Dixon, G. Fazekas, and G. A. Wiggins, “ATEPP: A dataset of automatically transcribed expressive piano performance,” inProceed- ings of the 23rd International Society for Music Infor- mation Retrieval Conference (ISMIR), Bengaluru,India, 2022,...

    2022

  17. [25]

    ASAP: A dataset of aligned scores and per- formances for piano transcription,

    F. Foscarin, A. McLeod, P. Rigaux, F. Jacquemard, and M. Sakai, “ASAP: A dataset of aligned scores and per- formances for piano transcription,” inProceedings of the 21st International Society for Music Information Retrieval Conference (ISMIR), 2020, pp. 534–541

    2020

  18. [26]

    A path towards autonomous machine intelli- gence,

    Y . LeCun, “A path towards autonomous machine intelli- gence,”Open Review, vol. 62, no. 1, pp. 1–62, 2022

    2022

  19. [27]

    LLM-JEPA: Large language models meet joint embedding predictive architectures,

    H. Huang, Y . LeCun, and R. Balestriero, “LLM-JEPA: Large language models meet joint embedding predictive architectures,” inProceedings of the 14th International Conference on Learning Representations (ICLR), 2026

    2026

  20. [28]

    Pop music transformer: Beat-based modeling and generation of expressive pop piano compositions,

    Y .-S. Huang and Y .-H. Yang, “Pop music transformer: Beat-based modeling and generation of expressive pop piano compositions,” inProceedings of the 28th ACM International Conference on Multimedia, ser. MM ’20, 2020, pp. 1180–1188

    2020

  21. [29]

    Compound word transformer: Learning to compose full-song music over dynamic directed hypergraphs,

    W.-Y . Hsiao, J.-Y . Liu, and Y .-H. Yang, “Compound word transformer: Learning to compose full-song music over dynamic directed hypergraphs,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 35, no. 1, 2021, pp. 178–186

    2021

  22. [30]

    MusicBERT: Symbolic music understanding with large- scale pre-training,

    M. Zeng, X. Tan, R. Wang, Z. Ju, T. Qin, and T.-Y . Liu, “MusicBERT: Symbolic music understanding with large- scale pre-training,” inFindings of the Association for Computational Linguistics: ACL-IJCNLP 2021, 2021, pp. 791–800

    2021

  23. [31]

    PerTok: Expressive encoding and modeling of symbolic musical ideas and variations,

    J. Lenz and A. Mani, “PerTok: Expressive encoding and modeling of symbolic musical ideas and variations,” in Proceedings of the 25th International Society for Music Information Retrieval Conference (ISMIR), 2024, pp. 981–988

    2024

  24. [32]

    MidiBERT-Piano: Large-scale pre-training for symbolic music understanding,

    Y .-H. Chou, I.-C. Chen, C.-J. Chang, J. Ching, and Y .- H. Yang, “MidiBERT-Piano: Large-scale pre-training for symbolic music understanding,”Journal of Creative Music Systems (JCMS), vol. 8, no. 1, 2024

    2024

  25. [33]

    PianoBART: Symbolic piano music gen- eration and understanding with large-scale pre-training,

    X. Liang, Z. Zhao, W. Zeng, Y . He, F. He, Y . Wang, and C. Gao, “PianoBART: Symbolic piano music gen- eration and understanding with large-scale pre-training,” in2024 IEEE International Conference on Multimedia and Expo (ICME), 2024, pp. 1–6

    2024

  26. [34]

    Chat- Musician: Understanding and generating music intrin- sically with LLM,

    R. Yuan, H. Lin, Y . Wang, Z. Tian, S. Wu, T. Shen, G. Zhang, Y . Wu, C. Liu, Z. Zhou, Z. Ma, L. Xue, Z. Wang, Q. Liu, T. Zheng, Y . Li, Y . Ma, Y . Liang, X. Chi, R. Liu, Z. Wang, P. Li, J. Wu, C. Lin, Q. Liu, T. Jiang, W. Huang, W. Chen, E. Benetos, J. Fu, G. Xia, R. Dannenb...

    2024

  27. [35]

    MuseCoco: Generating symbolic music from text,

    P. Lu, X. Xu, C. Kang, B. Yu, C. Xing, X. Tan, and J. Bian, “MuseCoco: Generating symbolic music from text,”arXiv preprint arXiv:2306.00110, 2023

    2023 arXiv

  28. [36]

    Moonbeam: A midi foundation model using both absolute and relative music attributes,

    Z. Guo and S. Dixon, “Moonbeam: A midi foundation model using both absolute and relative music attributes,” arXiv preprint arXiv:2505.15559, 2025

    2025 arXiv

  29. [37]

    Scaling self-supervised representation learn- ing for symbolic piano performance,

    L. Bradshaw, H. Fan, A. Spangher, S. Biderman, and S. Colton, “Scaling self-supervised representation learn- ing for symbolic piano performance,” inProceedings of the 26th International Society for Music Information Retrieval Conference (ISMIR), 2025, pp. 451–459

    2025

  30. [38]

    CLaMP 3: Universal music information retrieval across unaligned modalities and unseen languages,

    S. Wu, Z. Guo, R. Yuan, J. Jiang, S. Doh, G. Xia, J. Nam, X. Li, F. Yu, and M. Sun, “CLaMP 3: Universal music information retrieval across unaligned modalities and unseen languages,” inFindings of the Association for Computational Linguistics: ACL 2025, 2025, pp. 2605– 2625

    2025

  31. [39]

    Pianist transformer: To- wards expressive piano performance rendering via scalable self-supervised pre-training,

    H.-J. You, J.-J. Shao, X.-W. Yang, L.-H. Jia, L.- Z. Guo, and Y .-F. Li, “Pianist transformer: To- wards expressive piano performance rendering via scalable self-supervised pre-training,”arXiv preprint arXiv:2512.02652, 2025

    2025 arXiv

  32. [40]

    Combining piano performance dimensions for score difficulty classification,

    P. Ramoneda, D. Jeong, V . Eremenko, N. C. Tamer, M. Miron, and X. Serra, “Combining piano performance dimensions for score difficulty classification,”Expert Systems with Applications, p. 121776, 2024

    2024

  33. [41]

    Pianist identifica- tion using convolutional neural networks,

    J. Tang, G. Wiggins, and G. Fazekas, “Pianist identifica- tion using convolutional neural networks,” inProceed- ings of the 4th International Symposium on the Internet of Sounds (IS2), 2023

    2023

  34. [42]

    Lad- dersym: A multimodal interleaved transformer for mu- sic practice error detection,

    B. S.-H. Chou, P. Jajal, N. J. Eliopoulos, J. C. Davis, G. K. Thiruvathukal, K. Y .-J. Yun, and Y .-H. Lu, “Lad- dersym: A multimodal interleaved transformer for mu- sic practice error detection,” inThe Fourteenth Interna- tional Conference on Learning Representations (ICLR), 2026

    2026

  35. [43]

    Self-supervised learning from images with a joint- embedding predictive architecture,

    M. Assran, Q. Duval, I. Misra, P. Bojanowski, P. Vincent, M. Rabbat, Y . LeCun, and N. Ballas, “Self-supervised learning from images with a joint- embedding predictive architecture,” in2023 IEEE/CVF Conference on Computer Vision and Pattern Recogni- tion (CVPR), 2023, pp. 15 6...

    2023

  36. [44]

    V-JEPA 2: Self-supervised video models enable understanding, prediction and planning,

    M. Assran, A. Bardes, D. Fan, Q. Garrido, R. Howes, M. Komeili, M. Muckley, A. Rizvi, C. Roberts, K. Sinha, A. Zholus, S. Arnaud, A. Gejji, A. Martin, F. R. Hogan, D. Dugas, P. Bojanowski, V . Khalidov, P. Labatut, F. Massa, M. Szafraniec, K. Krishnakumar, Y . Li, X. Ma, S. Ch...

    2025 arXiv

  37. [45]

    Audio-JEPA: Joint-embedding predictive architecture for audio representation learning,

    L. Tuncay, E. Labbé, E. Benetos, and T. Pellegrini, “Audio-JEPA: Joint-embedding predictive architecture for audio representation learning,” inICME 2025, Nantes, France, 2025

    2025

  38. [46]

    Stem-JEPA: A joint-embedding predictive architecture for musical stem compatibility estimation,

    A. Riou, S. Lattner, G. Hadjeres, M. Anslow, and G. Peeters, “Stem-JEPA: A joint-embedding predictive architecture for musical stem compatibility estimation,” inProceedings of the 25th International Society for Mu- sic Information Retrieval Conference (ISMIR), 2024

    2024

  39. [47]

    Compose with me: Collaborative music inpainter for symbolic music infilling,

    Z. Hu, Y . Liu, G. Chen, and B. X. B. Yu, “Compose with me: Collaborative music inpainter for symbolic music infilling,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 39, no. 2, 2025, pp. 1327– 1335

    2025

  40. [48]

    Self-supervised representation learning with a JEPA framework for multi-instrument music transcription,

    M. Pilataki, M. Mauch, and S. Dixon, “Self-supervised representation learning with a JEPA framework for multi-instrument music transcription,” inProceedings of the IEEE Workshop on Applications of Signal Pro- cessing to Audio and Acoustics (WASPAA), 2025, pp. 1–5

    2025

  41. [49]

    Using a joint-embedding predictive architecture for symbolic music understand- ing,

    R. Hachana and B. Rasheed, “Using a joint-embedding predictive architecture for symbolic music understand- ing,” inAI for Music Workshop, 2025

    2025

  42. [50]

    Scoring time intervals using non- hierarchical transformer for automatic piano transcrip- tion,

    Y . Yan and Z. Duan, “Scoring time intervals using non- hierarchical transformer for automatic piano transcrip- tion,” inProceedings of the 25th International Society for Music Information Retrieval Conference (ISMIR), 2024, pp. 973–980

    2024

  43. [51]

    Aria-amt: Efficient and robust automatic piano transcription,

    L. Bradshaw and EleutherAI, “Aria-amt: Efficient and robust automatic piano transcription,” 2024. [Online]. Available: https://github.com/EleutherAI/aria-amt

    2024

  44. [52]

    LoRA: Low-rank adaptation of large language models,

    E. J. Hu, Y . Shen, P. Wallis, Z. Allen-Zhu, Y . Li, S. Wang, L. Wang, and W. Chen, “LoRA: Low-rank adaptation of large language models,” inInternational Conference on Learning Representations (ICLR), 2022

    2022

  45. [53]

    RoFormer: Enhanced transformer with rotary posi- tion embedding,

    J. Su, Y . Lu, S. Pan, A. Murtadha, B. Wen, and Y . Liu, “RoFormer: Enhanced transformer with rotary posi- tion embedding,”Neurocomputing, vol. 568, p. 127063, 2024

    2024

  46. [54]

    Audio- based piano performance evaluation for beginners with convolutional neural network and attention mechanism,

    W. Wang, J. Pan, H. Yi, Z. Song, and M. Li, “Audio- based piano performance evaluation for beginners with convolutional neural network and attention mechanism,” IEEE/ACM Transactions on Audio Speech and Lan- guage Processing, vol. 29, pp. 1119–1133, 2021

    2021

  47. [55]

    Piano skills assess- ment,

    P. Parmar, J. Reddy, and B. Morris, “Piano skills assess- ment,” in2021 IEEE 23rd International Workshop on Multimedia Signal Processing (MMSP), 2021, pp. 1–5

    2021

  48. [56]

    Expert and novice evaluations of piano per- formances: Criteria for computer-aided feedback,

    Y . Jiang, “Expert and novice evaluations of piano per- formances: Criteria for computer-aided feedback,” in Proceedings of the 24th International Society for Music Information Retrieval (ISMIR), 2023, pp. 367–374

    2023

  49. [57]

    Score-informed transcription for automatic piano tutoring,

    E. Benetos, A. Klapuri, and S. Dixon, “Score-informed transcription for automatic piano tutoring,” inProceed- ings of the 20th European Signal Processing Conference (EUSIPCO), 2012, pp. 2153–2157

    2012

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.