Pith. sign in

REVIEW 3 major objections 5 minor 72 references

Learning Music Style for Piano Arrangement Through Cross-Modal Bootstrapping

T0 review · 3 major / 5 minor · reviewed 2026-08-08 · deepseek-v4-flash

Pith's one-line read A lightweight connector between a frozen audio model and a frozen symbolic music model lets a lead sheet and a reference recording jointly control a piano arrangement, transferring the recording's groove, dynamics, and tempo into symbolic…

desk verdict Worth a serious referee: the arrangement results are solid, but the retrieval experiment never says its pairs are held out, so the representation claim needs that clarified before it can be trusted. read the letter →

arxiv 2608.03050 v1 pith:DM5L26NA submitted 2026-08-04 cs.SD cs.AIcs.MMeess.AS

classification cs.SDcs.AIcs.MMeess.AS
keywords cross-modalmusicrepresentationpianocovergenerationimplicitstyletransferaudio-to-MIDIretrievalQ-Formercontrastivelearningsymbolic
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

This paper argues that music style—the groove, dynamics, and tempo that make a performance feel a certain way—can be pulled out of a raw audio recording and stamped onto a symbolic piano arrangement, without retraining either a large audio model or a large symbolic music model. The authors connect the two frozen models with a Querying Transformer (Q-Former), a small trainable module that learns a compressed style representation from the audio model's hidden states and feeds it, together with a lead sheet, into the symbolic model. Training happens in two stages: first, contrastive and matching objectives align the audio and symbolic sides and a generative objective makes the style representation usable; second, a lightweight adapter conditions the symbolic model on the extracted style. If the claim holds, composers and arrangers could control not just what notes are played but how they are played by choosing a reference recording. The paper reports evidence on piano cover generation, style transfer, and audio-to-MIDI retrieval, including out-of-distribution genres.

What carries the argument

The load-bearing component is the Q-Former, a Querying Transformer: a Transformer encoder with $K=32$ learnable query vectors that cross-attend to the hidden states of a frozen audio language model while sharing self-attention with a stream of symbolic piano tokens. At test time only the audio stream remains, producing a style embedding $\mathbf{Z}$ that conditions a frozen symbolic music language model through a linear projection and a LoRA adapter, alongside the lead sheet that supplies content. Stage-I trains the Q-Former with three masked objectives—audio-symbolic contrastive learning, audio-symbolic matching with hard negatives, and audio-grounded symbolic generation—each with a self-attention mask that controls how much the two modalities see each other. The design acts as a bottleneck: the queries pass style-related information forward while the lead sheet carries melody and harmony, so the large backbones never need to be retrained.

What would settle it

Build a test set of 4-bar audio–MIDI pairs in which the style clearly changes at the midpoint (for example, straight eighths switching to swing, or a sudden tempo and texture break), generate covers conditioned on these clips, and count whether the second half's groove appears in the symbolic output. If the output reflects only an average of the two styles, the local-consistency assumption that carries the method gives way.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that a frozen audio language model and a frozen symbolic music language model can be bridged by a small Q-Former trained to extract style rather than content, and that the resulting system produces piano covers that are more stylistically coherent—matching the reference audio's grooving patterns, velocity contours, and tempo—than existing audio-to-symbolic arrangement methods. The style signal survives deliberately loose data pairing: 10-second audio clips are matched to 4-bar MIDI segments with random temporal shifts up to one second and random transposition into all twelve keys, so the model cannot memorize note-for-note correspondence and must learn the locally consistent style. The same representation supports cross-modal style transfer (a lead sheet paired with an unrelated style reference) and, after Stage-I alone, audio-to-MIDI retrieval whose accuracy is nearly unchanged when all candidates are transposed, showing the learned alignment is key-invariant and style-based.

Load-bearing premise

The load-bearing premise is that style is consistent within each paired 10-second audio clip and 4-bar MIDI segment, so random temporal shifts and twelve-key transposition still leave the model a stable style signal to learn; if style shifts inside that window, or if the loose alignment throws away the timing cues that define the groove, the training signal degrades.

Editorial extensions

If this is right

  • Given a lead sheet and an audio reference from the same song, the system generates a piano cover whose groove, dynamics, and tempo track the reference, as measured by Grooving Pattern Coherence, Velocity Contour Coherence, and Tempo Accuracy.
  • Given a lead sheet and an unrelated audio reference, the same model performs audio-to-symbolic style transfer, e.g., rendering the same melody as ragtime, bossa nova, or orchestral ballad depending on the chosen reference.
  • The Q-Former alone, without the generative stage, can retrieve the matching MIDI segment for an audio query from a pool of 128 candidates, and its accuracy barely changes when candidates are transposed across keys, indicating style-based rather than pitch-based alignment.
  • Only the 186M-parameter Q-Former and a 5M-parameter LoRA adapter are trained; the billion-scale audio and symbolic backbones stay frozen, making the approach comparatively cheap to adapt to new styles.
  • The performance difference between the full two-stage training and a no-pre-training ablation is largest on multi-instrumental audio, suggesting the contrastive, matching, and generative objectives matter most when style must be separated from richer audio textures.

Reading between the lines

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

  • Not tested in the paper: the same Q-Former style embedding could be probed dimension-by-dimension to see which coordinates encode swing, tempo, or dynamics; if such separation exists, style editing in latent space becomes possible without new training.
  • Not tested in the paper: because the representation is transposition-invariant and trained on short segments, it likely discards the exact micro-timing that distinguishes a tight swing feel; a tightly aligned variant with onset-time supervision would show whether that precision can be recovered.
  • Not tested in the paper: the local-consistency assumption implies the model will average over a style change inside a 4-bar window; a hierarchical or temporally adaptive style model would be needed for long-form compositions with phrase-level expressive arcs.
  • Not tested in the paper: the retrieval result suggests the style representation might serve as a general cross-modal music similarity measure, applicable to tasks like cover-song identification or style-based music search.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper proposes a two-stage cross-modal framework for piano arrangement in which a Q-Former is trained to extract implicit style representations from a frozen audio language model (MusicGen) and to condition a frozen symbolic language model (MuseCoco) for generating piano performances from a lead sheet and a reference audio clip. Stage-I uses contrastive, matching, and audio-grounded generative losses on loosely aligned 10-second audio and 4-bar MIDI pairs; Stage-II fine-tunes a LoRA adapter while keeping the symbolic LM frozen. The paper evaluates the approach on piano cover generation, cross-modal style transfer, audio-to-MIDI retrieval, and a subjective listening test, reporting improvements in style coherence metrics over PiCoGen2 and an Audio-to-MIDI disentanglement baseline, as well as an ablation showing the value of Stage-I pre-training.

Significance. If the central claim holds, the paper offers a scalable way to transfer implicit musical style across audio and symbolic modalities without retraining large backbone models, which would be useful for controllable arrangement and style-conditioned generation. The architecture is described in detail, the two-stage training procedure is clear, and the authors provide a demo page and a candid limitations section. The main experimental strength is the combination of multiple objective metrics, an ablation study, and a user study. However, the cleanest evidence for a transferable style representation — the audio-to-MIDI retrieval experiment — currently has a potentially serious methodological gap, and the subjective evaluation relies on a small, non-blind sample-selection procedure. These issues need to be addressed before the paper's claims are fully supported.

major comments (3)
  1. [Section 5.2.1, Table 3] The retrieval test set is described only as "randomly sampled from PIAST and POP909 (64 pairs each)" and does not state that these pairs are drawn from the held-out test splits defined in Section 4.1. Because the Q-Former is trained on PIAST and POP909 with contrastive and matching losses, including training pairs would make the reported Acc@1 of 71.4% reflect memorization rather than a transferable cross-modal representation. The paper must state explicitly that the 128 pairs are disjoint from the training and validation sets, and if they are not, Table 3 should be re-run on the held-out test splits.
  2. [Section 5.2.1, Table 3] Even if the retrieval pairs are held out, the task pairs each audio query with the MIDI segment from the same song, so high retrieval accuracy can be achieved through content cues such as melody contour, harmony, and rhythm rather than through style. The random-transposition condition removes absolute pitch as a cue, but it does not remove relative melodic and harmonic content. Therefore the claim that the Q-Former captures "stylistic coherence beyond surface-level attributes" is not established by this experiment. I recommend adding a retrieval condition in which positive pairs share style but not content (for example, different songs from the same genre or artist), or otherwise explicitly controlling for content similarity.
  3. [Section 4.4 and Appendix C.2] The subjective evaluation selects the "best result from 3 generated samples" for each model without a pre-registered or blinded selection criterion. With only 21 participants and 3 test pieces per participant, the significant preference for "Ours" in Coherence and Musicality is not robust evidence of superiority. I recommend either reporting results across all generated samples, using a blind and pre-defined selection rule, or presenting a sensitivity analysis with random sample selection.
minor comments (5)
  1. [Between Section 4.3 and Figure 3] The manuscript contains an apparent rendering artifact — a long string of "/uni..." tokens — that should be removed or fixed before publication.
  2. [Appendix A.1] The text says the symbolic arrangements are tokenized in "OctMIDI format [52]" but reference [52] is the MusicBERT paper, which uses OctupleMIDI encoding; please check the terminology and citation.
  3. [Table 1 caption] The caption contains "p-valuep < 0.05/6", which should read "p-value p < 0.05/6" with appropriate spacing.
  4. [Section 4.4] The study is described as "double-blind," but participants are only blind to model identity while the authors select the best samples; please clarify the exact blinding procedure.
  5. [Section 5.2.1] The description of "10 independent runs" should clarify whether the same audio and MIDI segments may appear in multiple runs, and whether the 128 candidates are unique within each run.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: the central audio-to-symbolic style derivation is self-contained and evaluated on held-out and out-of-distribution data; the only self-citations are peripheral design choices, not load-bearing.

full rationale

The paper's derivation chain is not circular. Stage-I trains a Q-Former with contrastive, matching, and generative losses on loosely aligned audio-MIDI pairs; Stage-II conditions a frozen MuseCoco with the resulting style embedding plus a lead sheet. No equation or construction in the paper reduces a predicted quantity to a fitted input: the style representation is evaluated on the POP909 test split, on held-out PIAST test audio for style transfer, and on out-of-distribution Ballroom/GTZAN tracks. The GPC, VCC, and TA metrics are either explicitly cited from prior work or defined directly from ground-truth annotations, and the subjective study is an independent human rating. The paper contains minor self-citations ([48] for chord accuracy and [7] for the choice of MusicGen layer), but neither is load-bearing for the central claim. A real validity concern remains: Section 5.2.1 does not explicitly state that the 128 retrieval pairs are drawn from held-out test splits, so if they overlapped training data the Acc@1 result would be inflated by memorization. However, that is a data-hygiene risk that cannot be established from the text, not a circular derivation, and the paper's main generation results stand on held-out and OOD evaluations.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central claim rests on a data-pairing assumption about local style consistency and on the validity of the proposed evaluation metrics; both are stated clearly but are not independently verified. The model itself contributes many learned parameters, while the hand-chosen hyperparameters listed above affect the reported numbers.

free parameters (4)
  • Q-Former query count K = 32
    Number of learnable query embeddings in the Q-Former bottleneck, chosen by hand following BLIP-2 (Appendix A.1).
  • LoRA rank = 16
    Rank of the LoRA adapter inserted into MuseCoco self-attention layers (Appendix A.3).
  • Top-k sampling k = 15
    Decoding hyperparameter at test time (Appendix A.4).
  • Tempo accuracy tolerance = 0.08
    Relative tempo tolerance in Eq. 2 of Appendix B.5, adopted from mir_eval; it directly affects the reported TA values.
assumptions (5)
  • domain assumption Music style is locally consistent at the segment or bar level (4-bar windows).
    Section 3.1 pairs 10s audio with 4-bar MIDI assuming style is consistent within the segment; Appendix E explicitly acknowledges this assumption and its limitation for longer phrases.
  • domain assumption Frozen pre-trained LMs (MusicGen and MuseCoco) contain sufficient style and content information without fine-tuning.
    Section 3.4 keeps both backbones frozen and only trains the Q-Former and a LoRA adapter, relying on their internal representations.
  • ad hoc to paper Random temporal shift and random transposition in data pairing prevent content memorization while preserving style cues.
    Section 3.1 introduces plus or minus 1 second shift and 12-key transposition to discourage note-to-note correspondence; this is a paper-specific design choice rather than a standard prior.
  • domain assumption The objective metrics (MCA, CA, GPC, VCC, TA) are valid proxies for content preservation and style coherence.
    Section 4.3 and Appendix B define metrics, some from prior work and two (VCC, TA) with original formulas; the paper assumes these capture the perceptual feel of the performance.
  • domain assumption Sheetsage-transcribed lead sheets are sufficiently accurate for all compared methods.
    Section 4.2 uses Sheetsage to extract lead sheets from audio as the sole content input, so errors propagate to all models equally but could lower content scores.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Learning Music Style for Piano Arrangement Through Cross-Modal Bootstrapping." pith.science (2026). https://pith.science/paper/DM5L26NA

@misc{pith2026260803050,
  author       = {Pith},
  title        = {Pith review of: Learning Music Style for Piano Arrangement Through Cross-Modal Bootstrapping},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DM5L26NA}},
  note         = {Machine review of arXiv:2608.03050}
}
read the original abstract

What is music style? Though often described using text labels such as "swing," "classical," or "emotional," the real style remains implicit and hidden in concrete music examples. In this paper, we introduce a cross-modal framework that learns implicit music styles from raw audio and applies them to symbolic music generation. Inspired by BLIP-2, our model leverages a Querying Transformer (Q-Former) to extract style representations from a large, pre-trained audio language model (LM), and further applies them to condition a symbolic LM for generating piano arrangements. We adopt a two-stage training strategy: contrastive learning to align auditory style with symbolic expression, followed by generative modeling for music arrangement. Our model generates piano performances jointly conditioned on a lead sheet (content) and a reference audio example (style), enabling controllable and stylistically faithful arrangement. Experiments demonstrate the effectiveness of our approach in piano cover generation, style transfer, and audio-to-MIDI retrieval, achieving substantial improvements in style-aware alignment and music quality.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

72 extracted references · 64 canonical work pages

  1. [1]

    Learning Music Style for Piano Arrangement Through Cross-Modal Bootstrapping

    INTRODUCTION Automatic music generation is often controlled byexplicit content such as melody, chords, and text labels [1–4], but music concepts can be more nuanced than we often real- ize. When musicians learn a style, instead of relying on abstract descriptors like “romantic” or “jazz” alone, they absorb patterns from music examples that share common st...

  2. [2]

    We use Q-Former toalign audio and symbolic modalities via implicit music style, extending its role beyond content alignment in vision-language tasks

  3. [3]

    We presenta new methodology to disentangle music style from large, pre-trained LMs, offering a more scalable alternative to traditional latent-variable dis- entanglement methods

  4. [4]

    Experiments demonstrate that it outperforms existing audio-to-symbolic mod- els, including both disentanglement-based methods and standard LM approaches

    Our model achievesstyle-aware audio-to-symbolic piano cover arrangement. Experiments demonstrate that it outperforms existing audio-to-symbolic mod- els, including both disentanglement-based methods and standard LM approaches

  5. [5]

    Section 2.1 overviews re- cent advances in music LMs, while Section 2.2 focuses on piano cover generation, a primary task of this paper

    RELATED WORK We review two relevant areas. Section 2.1 overviews re- cent advances in music LMs, while Section 2.2 focuses on piano cover generation, a primary task of this paper. 2.1 Music Language Models Rapid progress in large-scale language models has trans- formed how we interact with various forms of media, in- cluding text, image, and music [16–20]...

  6. [6]

    In Section 3.1, we first introduce our audio-symbolic data pairing method that fa- cilitates style learning

    METHOD To bridge the modality gap from audio to symbolic mu- sic, we adopt the Q-Former [16] under a two-stage training strategy, as shown in Figure 2. In Section 3.1, we first introduce our audio-symbolic data pairing method that fa- cilitates style learning. We illustrate the Q-Former archi- tecture in Section 3.2, followed by the two-stage training pro...

  7. [7]

    When the two inputs are aligned with each other, the task corre- sponds topiano cover generation; when they are unpaired, the task becomescross-modal style transfer

    EXPERIMENTS Our model generates piano performances jointly condi- tioned on a lead sheet and an audio reference. When the two inputs are aligned with each other, the task corre- sponds topiano cover generation; when they are unpaired, the task becomescross-modal style transfer. This section focuses onpiano cover generation, which allows direct comparison ...

  8. [8]

    Specifically, we examinecross-modal style transferin Sec- tion 5.1, and audio-to-symbolic retrieval in Section 5.2

    ADDITIONAL EV ALUATIONS In this section, we explore additional experimental set- tings to further evaluate our model’s capabilities, with a particular focus on cross-modal representation learning. Specifically, we examinecross-modal style transferin Sec- tion 5.1, and audio-to-symbolic retrieval in Section 5.2. 5.1 Evaluation on Cross-Modal Style Transfer...

Show all 72 references
  1. [9]

    CONCLUSION In this paper, we introduce a cross-modal framework for audio-to-symbolic arrangement. By re-purposing the Q- Former to align audio and symbolic modalities, our model extracts and applies implicit music style using pre-trained music LMs, enabling expressive piano ar...

  2. [10]

    Deep music analogy via latent representation disentanglement,

    R. Yang, D. Wang, Z. Wang, T. Chen, J. Jiang, and G. Xia, “Deep music analogy via latent representation disentanglement,” inProceedings of the 20th Interna- tional Society for Music Information Retrieval Confer- ence, ISMIR 2019, 2019, pp. 596–603

  3. [11]

    Learning in- terpretable representation for controllable polyphonic music generation,

    Z. Wang, D. Wang, Y . Zhang, and G. Xia, “Learning in- terpretable representation for controllable polyphonic music generation,” inProceedings of the 21st Interna- tional Society for Music Information Retrieval Confer- ence, ISMIR 2020, 2020, pp. 662–669

  4. [12]

    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

  5. [13]

    Text2midi: Generating symbolic mu- sic from captions,

    K. Bhandari, A. Roy, K. Wang, G. Puri, S. Colton, and D. Herremans, “Text2midi: Generating symbolic mu- sic from captions,” inAAAI-25, Sponsored by the Asso- ciation for the Advancement of Artificial Intelligence. AAAI Press, 2025, pp. 23 478–23 486

  6. [14]

    Do music generation models encode music theory?

    M. Wei, M. Freeman, C. Donahue, and C. Sun, “Do music generation models encode music theory?” in Proceedings of the 25th International Society for Mu- sic Information Retrieval Conference, ISMIR 2024, 2024, pp. 680–687

  7. [15]

    Exploring the internal mechanisms of music llms: A study of root and quality via probing and intervention techniques,

    W. Ma and G. Xia, “Exploring the internal mechanisms of music llms: A study of root and quality via probing and intervention techniques,” inICML 2024 Workshop on Mechanistic Interpretability, 2024

  8. [16]

    Do music llms learn sym- bolic concepts? a pilot study using probing and inter- vention,

    W. Ma, X. Li, and G. Xia, “Do music llms learn sym- bolic concepts? a pilot study using probing and inter- vention,” inAudio Imagination: NeurIPS 2024 Work- shop AI-Driven Speech, Music, and Sound Generation, 2024

  9. [17]

    Exploring the inner mechanisms of large generative music models,

    M. A. V . Vásquez, C. Pouw, J. A. Burgoyne, and W. H. Zuidema, “Exploring the inner mechanisms of large generative music models,” inProceedings of the 25th International Society for Music Information Retrieval Conference, ISMIR 2024, 2024, pp. 791–798

  10. [18]

    Codified au- dio language modeling learns useful representations for music information retrieval,

    R. Castellon, C. Donahue, and P. Liang, “Codified au- dio language modeling learns useful representations for music information retrieval,” inProceedings of the 22nd International Society for Music Information Re- trieval Conference, ISMIR 2021, 2021, pp. 88–96

  11. [19]

    Content-based controls for music large language modeling,

    L. Lin, G. Xia, J. Jiang, and Y . Zhang, “Content-based controls for music large language modeling,” inPro- ceedings of the 25th International Society for Music Information Retrieval Conference, ISMIR 2024, 2024, pp. 783–790

  12. [20]

    Instruct-musicgen: Unlocking text-to-music editing for music language models via instruction tuning,

    Y . Zhang, Y . Ikemiya, W. Choi, N. Murata, M. A. Martínez-Ramírez, L. Lin, G. Xia, W.-H. Liao, Y . Mit- sufuji, and S. Dixon, “Instruct-musicgen: Unlocking text-to-music editing for music language models via instruction tuning,”arXiv preprint arXiv:2405.18386, 2024

  13. [21]

    Arrange, in- paint, and refine: Steerable long-term music audio gen- eration and editing via content-based controls,

    L. Lin, G. Xia, Y . Zhang, and J. Jiang, “Arrange, in- paint, and refine: Steerable long-term music audio gen- eration and editing via content-based controls,” inPro- ceedings of the Thirty-Third International Joint Con- ference on Artificial Intelligence, IJCAI 2024. ij- cai....

  14. [22]

    Music controlnet: Multiple time-varying controls for music generation,

    S.-L. Wu, C. Donahue, S. Watanabe, and N. J. Bryan, “Music controlnet: Multiple time-varying controls for music generation,”IEEE/ACM Transactions on Audio, Speech, and Language Processing, vol. 32, pp. 2692– 2703, 2024

  15. [23]

    Melody tran- scription via generative pre-training,

    C. Donahue, J. Thickstun, and P. Liang, “Melody tran- scription via generative pre-training,” inProceedings of the 23rd International Society for Music Information Retrieval Conference, ISMIR 2022, 2022, pp. 485–492

  16. [24]

    Audio-to- symbolic arrangement via cross-modal music repre- sentation learning,

    Z. Wang, D. Xu, G. Xia, and Y . Shan, “Audio-to- symbolic arrangement via cross-modal music repre- sentation learning,” inIEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP

  17. [25]

    High- resolution piano transcription with pedals by regress- ing onset and offset times,

    Q. Kong, B. Li, X. Song, Y . Wan, and Y . Wang, “High- resolution piano transcription with pedals by regress- ing onset and offset times,”IEEE ACM Trans. Audio Speech Lang. Process., vol. 29, pp. 3707–3717, 2021

  18. [26]

    BLIP-2: bootstrapping language-image pre-training with frozen image encoders and large language models,

    J. Li, D. Li, S. Savarese, and S. C. H. Hoi, “BLIP-2: bootstrapping language-image pre-training with frozen image encoders and large language models,” inIn- ternational Conference on Machine Learning, ICML 2023, ser. Proceedings of Machine Learning Research, vol. 202. PMLR, 20...

  19. [27]

    Llama: Open and efficient foundation language models,

    H. Touvron, T. Lavril, G. Izacard, X. Martinet, M.-A. Lachaux, T. Lacroix, B. Rozière, N. Goyal, E. Hambro, F. Azharet al., “Llama: Open and efficient foundation language models,”arXiv preprint arXiv:2302.13971, 2023

  20. [28]

    Flamingo: a visual lan- guage model for few-shot learning,

    J. Alayrac, J. Donahue, P. Luc, A. Miech, I. Barr, Y . Hasson, K. Lenc, A. Mensch, K. Millican, M. Reynolds, R. Ring, E. Rutherford, S. Cabi, T. Han, Z. Gong, S. Samangooei, M. Monteiro, J. L. Menick, S. Borgeaud, A. Brock, A. Nematzadeh, S. Shar- ifzadeh, M. Binkowski, R. Bar...

  21. [29]

    Audio flamingo: A novel audio lan- guage model with few-shot learning and dialogue abil- ities,

    Z. Kong, A. Goel, R. Badlani, W. Ping, R. Valle, and B. Catanzaro, “Audio flamingo: A novel audio lan- guage model with few-shot learning and dialogue abil- ities,” inForty-first International Conference on Ma- chine Learning, ICML 2024, Vienna, Austria, July 21- 27, 2024. Ope...

  22. [30]

    Yue: Scaling open foundation models for long-form music genera- tion,

    R. Yuan, H. Lin, S. Guo, G. Zhang, J. Pan, Y . Zang, H. Liu, Y . Liang, W. Ma, X. Duet al., “Yue: Scaling open foundation models for long-form music genera- tion,”arXiv preprint arXiv:2503.08638, 2025

  23. [31]

    Musiclm: Generating music from text,

    A. Agostinelli, T. I. Denk, Z. Borsos, J. Engel, M. Verzetti, A. Caillon, Q. Huang, A. Jansen, A. Roberts, M. Tagliasacchiet al., “Musiclm: Generating music from text,”arXiv preprint arXiv:2301.11325, 2023

  24. [32]

    Mustango: Toward con- trollable text-to-music generation,

    J. Melechovský, Z. Guo, D. Ghosal, N. Majumder, D. Herremans, and S. Poria, “Mustango: Toward con- trollable text-to-music generation,” inProceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Hu- man Language Technol...

  25. [33]

    Anticipatory music transformer,

    J. Thickstun, D. L. W. Hall, C. Donahue, and P. Liang, “Anticipatory music transformer,”Transactions on Machine Learning Research, 2024

  26. [34]

    Simple and control- lable music generation,

    J. Copet, F. Kreuk, I. Gat, T. Remez, D. Kant, G. Syn- naeve, Y . Adi, and A. Défossez, “Simple and control- lable music generation,” inAdvances in Neural Infor- mation Processing Systems 36: Annual Conference on Neural Information Processing Systems 2023, NeurIPS 2023, 2023

  27. [35]

    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,” inThe Tenth In- ternational Conference on Learning Representations, ICLR 2022. OpenReview.net, 2022

  28. [36]

    Exploring transformer’s potential on automatic pi- ano transcription,

    L. Ou, Z. Guo, E. Benetos, J. Han, and Y . Wang, “Exploring transformer’s potential on automatic pi- ano transcription,” inIEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP

  29. [37]

    IEEE, 2022, pp. 776–780

  30. [38]

    MT3: multi-task multitrack music tran- scription,

    J. Gardner, I. Simon, E. Manilow, C. Hawthorne, and J. H. Engel, “MT3: multi-task multitrack music tran- scription,” inThe Tenth International Conference on Learning Representations, ICLR 2022. OpenRe- view.net, 2022

  31. [39]

    Automatic lyric transcription and automatic music transcription from multimodal singing,

    X. Gu, L. Ou, W. Zeng, J. Zhang, N. Wong, and Y . Wang, “Automatic lyric transcription and automatic music transcription from multimodal singing,”ACM Trans. Multim. Comput. Commun. Appl., vol. 20, no. 7, pp. 209:1–209:29, 2024

  32. [40]

    End-to-end real-world polyphonic piano audio-to-score transcription with hi- erarchical decoding,

    W. Zeng, X. He, and Y . Wang, “End-to-end real-world polyphonic piano audio-to-score transcription with hi- erarchical decoding,” inProceedings of the Thirty- Third International Joint Conference on Artificial In- telligence, IJCAI 2024. ijcai.org, 2024, pp. 7788– 7795

  33. [41]

    Statistical piano re- duction controlling performance difficulty,

    E. Nakamura and K. Yoshii, “Statistical piano re- duction controlling performance difficulty,”APSIPA Transactions on Signal and Information Processing, vol. 7, 2018

  34. [42]

    Picogen: Generate pi- ano covers with a two-stage approach,

    C. Tan, S. Guan, and Y . Yang, “Picogen: Generate pi- ano covers with a two-stage approach,” inProceedings of the 2024 International Conference on Multimedia Retrieval, ICMR 2024. ACM, 2024, pp. 1180–1184

  35. [43]

    Pico- gen2: Piano cover generation with transfer learning ap- proach and weakly aligned data,

    C. Tan, H. Ai, Y . Chang, S. Guan, and Y . Yang, “Pico- gen2: Piano cover generation with transfer learning ap- proach and weakly aligned data,” inProceedings of the 25th International Society for Music Information Re- trieval Conference, ISMIR 2024, 2024, pp. 555–562

  36. [44]

    Pop2piano : Pop audio-based piano cover generation,

    J. Choi and K. Lee, “Pop2piano : Pop audio-based piano cover generation,” inIEEE International Con- ference on Acoustics, Speech and Signal Processing ICASSP 2023. IEEE, 2023, pp. 1–5

  37. [45]

    Align before fuse: Vision and lan- guage representation learning with momentum distil- lation,

    J. Li, R. R. Selvaraju, A. Gotmare, S. R. Joty, C. Xiong, and S. C. Hoi, “Align before fuse: Vision and lan- guage representation learning with momentum distil- lation,” inAdvances in Neural Information Processing Systems 34: Annual Conference on Neural Informa- tion Processin...

  38. [46]

    On- sets and frames: Dual-objective piano transcription,

    C. Hawthorne, E. Elsen, J. Song, A. Roberts, I. Si- mon, C. Raffel, J. H. Engel, S. Oore, and D. Eck, “On- sets and frames: Dual-objective piano transcription,” inProceedings of the 19th International Society for Music Information Retrieval Conference, ISMIR 2018, 2018, pp. 50–57

  39. [47]

    POP909: A pop-song dataset for music arrangement generation,

    Z. Wang, K. Chen, J. Jiang, Y . Zhang, M. Xu, S. Dai, and G. Xia, “POP909: A pop-song dataset for music arrangement generation,” inProceedings of the 21st International Society for Music Information Retrieval Conference, ISMIR 2020, 2020, pp. 38–45

  40. [48]

    Piast: A multimodal piano dataset with audio, symbolic and text,

    H. Bang, E. Choi, M. Finch, S. Doh, S. Lee, G.-H. Lee, and J. Nam, “Piast: A multimodal piano dataset with audio, symbolic and text,” inProceedings of the 3rd Workshop on NLP for Music and Audio (NLP4MusA), 2024, pp. 5–10

  41. [49]

    An experimental compari- son of audio tempo induction algorithms,

    F. Gouyon, A. Klapuri, S. Dixon, M. Alonso, G. Tzane- takis, C. Uhle, and P. Cano, “An experimental compari- son of audio tempo induction algorithms,”IEEE Trans. Speech Audio Process., vol. 14, no. 5, pp. 1832–1844, 2006

  42. [50]

    Rhythmic pattern modeling for beat and downbeat tracking in musical audio,

    F. Krebs, S. Böck, and G. Widmer, “Rhythmic pattern modeling for beat and downbeat tracking in musical audio,” inProceedings of the 14th International Soci- ety for Music Information Retrieval Conference, ISMIR 2013, 2013, pp. 227–232

  43. [51]

    Musical genre classi- fication of audio signals,

    G. Tzanetakis and P. R. Cook, “Musical genre classi- fication of audio signals,”IEEE Trans. Speech Audio Process., vol. 10, no. 5, pp. 293–302, 2002

  44. [52]

    Swing ratio estimation,

    U. Marchand and G. Peeters, “Swing ratio estimation,” inProceedings of the 18th International Conference on Digital Audio Effects, DAFx-15, 2015, pp. 1–6

  45. [53]

    Sheet sage: Lead sheets from music audio,

    C. Donahue and P. Liang, “Sheet sage: Lead sheets from music audio,”ISMIR 2021 Late-Breaking and Demo, 2021

  46. [54]

    Jukebox: A generative model for music,

    P. Dhariwal, H. Jun, C. Payne, J. W. Kim, A. Radford, and I. Sutskever, “Jukebox: A generative model for music,”arXiv preprint arXiv:2005.00341, 2020

  47. [55]

    Deconstruct, analyse, reconstruct: How to improve tempo, beat, and down- beat estimation,

    S. Böck and M. E. P. Davies, “Deconstruct, analyse, reconstruct: How to improve tempo, beat, and down- beat estimation,” inProceedings of the 21th Interna- tional Society for Music Information Retrieval Confer- ence, ISMIR 2020, 2020, pp. 574–582

  48. [56]

    Large- vocabulary chord transcription via chord structure de- composition,

    J. Jiang, K. Chen, W. Li, and G. Xia, “Large- vocabulary chord transcription via chord structure de- composition,” inProceedings of the 20th International Society for Music Information Retrieval Conference, 2019, pp. 644–651

  49. [57]

    PYIN: A fundamental fre- quency estimator using probabilistic threshold distribu- tions,

    M. Mauch and S. Dixon, “PYIN: A fundamental fre- quency estimator using probabilistic threshold distribu- tions,” inIEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP 2014. IEEE, 2014, pp. 659–663

  50. [58]

    Popmag: Pop music accompaniment generation,

    Y . Ren, J. He, X. Tan, T. Qin, Z. Zhao, and T.-Y . Liu, “Popmag: Pop music accompaniment generation,” in Proceedings of the 28th ACM International Conference on Multimedia, 2020, pp. 1198–1206

  51. [59]

    Structured multi-track accompaniment arrangement via style prior modelling,

    J. Zhao, G. Xia, Z. Wang, and Y . Wang, “Structured multi-track accompaniment arrangement via style prior modelling,” inAdvances in Neural Information Pro- cessing Systems 38: Annual Conference on Neural In- formation Processing Systems 2024, NeurIPS 2024, 2024

  52. [60]

    The jazz transformer on the front line: Exploring the shortcomings of ai-composed mu- sic through quantitative measures,

    S. Wu and Y . Yang, “The jazz transformer on the front line: Exploring the shortcomings of ai-composed mu- sic through quantitative measures,” inProceedings of the 21st International Society for Music Information Retrieval Conference, 2020, pp. 142–149

  53. [61]

    Scheffe,The analysis of variance

    H. Scheffe,The analysis of variance. John Wiley & Sons, 1999, vol. 72

  54. [62]

    Clamp 3: Universal music information retrieval across unaligned modali- ties 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 modali- ties and unseen languages,” inFindings of the Associa- tion for Computational Linguistics, ACL 2025. Asso- ciation for C...

  55. [63]

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

    M. Zeng, X. Tan, R. Wang, Z. Ju, T. Qin, and T. Liu, “Musicbert: Symbolic music understanding with large- scale pre-training,” inFindings of the Association for Computational Linguistics: ACL/IJCNLP 2021, ser. Findings of ACL, vol. ACL/IJCNLP 2021. Associa- tion for Computatio...

  56. [64]

    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,” inMM ’20: The 28th ACM Inter- national Conference on Multimedia, 2020, pp. 1180– 1188

  57. [65]

    Decoupled weight de- cay regularization,

    I. Loshchilov and F. Hutter, “Decoupled weight de- cay regularization,” in7th International Conference on Learning Representations, ICLR 2019. OpenRe- view.net, 2019

  58. [66]

    Mir_eval: A transparent implementation of common MIR metrics,

    C. Raffel, B. McFee, E. J. Humphrey, J. Salamon, O. Nieto, D. Liang, and D. P. W. Ellis, “Mir_eval: A transparent implementation of common MIR metrics,” inProceedings of the 15th International Society for Music Information Retrieval Conference, ISMIR 2014, 2014, pp. 367–372

  59. [67]

    Hybrid trans- formers for music source separation,

    S. Rouard, F. Massa, and A. Défossez, “Hybrid trans- formers for music source separation,” inIEEE Inter- national Conference on Acoustics, Speech and Signal Processing ICASSP 2023. IEEE, 2023, pp. 1–5

  60. [69]

    librosa: Audio and music signal analysis in python,

    B. McFee, C. Raffel, D. Liang, D. P. W. Ellis, M. McVicar, E. Battenberg, and O. Nieto, “librosa: Audio and music signal analysis in python,” inPro- ceedings of the 14th Python in Science Conference 2015 (SciPy 2015). scipy.org, 2015, pp. 18–24

  61. [70]

    Manipulation of music for melody matching,

    A. L. Uitdenbogerd and J. Zobel, “Manipulation of music for melody matching,” inProceedings of the 6th ACM International Conference on Multimedia ’98. ACM, 1998, pp. 235–240

  62. [71]

    MIDI Chord Recognition via Bar- Level Modeling,

    J. Jiang, “MIDI Chord Recognition via Bar- Level Modeling,” https://github.com/music-x-lab/ midi-chord-recognition, 2025

  63. [72]

    Automatic extraction of tempo and beat from expressive performances,

    S. Dixon, “Automatic extraction of tempo and beat from expressive performances,”Journal of New Music Research, vol. 30, no. 1, pp. 39–58, 2001. A. MODEL AND TRAINING DETAILS Our model comprises three components: an audio LM, a symbolic music LM, and a Q-Former connecting the t...

  64. [2022]

    IEEE, 2022, pp. 181–185

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.