Pith. sign in

REVIEW 5 major objections 5 minor 38 references

Exploring Adapter Design Tradeoffs for Low Resource Music Generation

T0 review · 5 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read Mid-sized 40M adapters at late layers give the best quality-per-cost for low-resource music generation.

desk verdict Useful systematic adapter comparison for low-resource music PEFT, but the 40M sweet spot rests on single-run metrics. read the letter →

arxiv 2506.21298 v2 pith:GJOG44Y5 submitted 2025-06-26 cs.SD cs.AIcs.CLcs.LGcs.MMeess.AS

Atharva Mehta , Shivam Chauhan , Monojit Choudhury This is my paper
classification cs.SDcs.AIcs.CLcs.LGcs.MMeess.AS
keywords parameter-efficientfine-tuningadapter-basedlearningmusicgenerationHindustaniClassicalTurkishMakamGenMustangoFréchetAudioDistance
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 asks how to adapt two large music-generation models, MusicGen and Mustango, to two underrepresented genres, Hindustani Classical and Turkish Makam, when training data and compute are limited. It claims that adapter-based fine-tuning has a reliable design sweet spot: adapters of about 40 million parameters, inserted at the end of late model blocks, produce the best balance of audio quality and training cost, while smaller adapters lack capacity and larger ones overfit or distort the base model. It further claims that the best adapter architecture is genre-dependent: convolution-based adapters capture local ornamentations and work best for Hindustani Classical, while transformer-based adapters preserve long-range structure and work best for Turkish Makam. The result matters because it gives practitioners a concrete recipe, around 20 to 45 GPU hours, for bringing non-Western genres into existing music-generation models without full fine-tuning.

What carries the argument

The load-bearing mechanism is the adapter module: a small trainable block (a fully connected bottleneck, a CNN with dilated residual blocks and squeeze-and-excitation, or a transformer with multi-head self-attention) inserted into a frozen base model. The paper varies three design knobs, architecture, placement, and parameter scale from about 2M to 70M, and measures the effects with FAD and FD against a fixed 400-clip reference corpus. The 40M sweet spot emerges from two opposing forces: too few parameters cannot capture raga- or makam-specific structure, and too many parameters overfit the limited genre data and destabilize pre-trained representations; late-layer placement avoids corrupting the core musical features encoded in earlier layers, which middle-layer insertion destroys.

What would settle it

Re-run the size sweep (2M, 8M, 20M, 40M, 70M) for MusicGen-CNN and Mustango-CNN on both genres with at least five random seeds, computing FAD/FD with bootstrap confidence intervals over the 400-prompt corpus; if the 40M or 20M-40M confidence intervals overlap with the 20M and 70M intervals for either genre, the claimed sweet spot is not supported.

Watch

Extended reading notes

Core claim

The paper's central discovery is that adapter-based fine-tuning of music-generation models has a reliable design optimum: adapters of about 40 million parameters, placed at the end of late model blocks, give the best balance between audio quality and training cost, and this pattern holds for both base models studied. At 2M-8M parameters adapters lack the capacity to represent genre-specific melodic and rhythmic structure; at 70M they overfit the limited genre data and degrade fidelity, with the sharpest failure in MusicGen's linear adapters. Placement is equally decisive: inserting adapters in middle layers destroys generation entirely (beeping, static, no musical structure), while late-layer or block-end placement preserves the base model's core musical representations and lets the adapter add stylistic detail. Architecture preference is genre-dependent: convolution-based adapters win for Hindustani Classical, where ornamentations such as gamakas, meends, and murkis dominate, and transformer-based adapters win for Turkish Makam, where long modal structure and phrase development matter. The paper also finds that the diffusion-based Mustango produces more diverse and prompt-faithful audio but suffers from note instability, rhythm misalignment, and higher training cost, while the autoregressive MusicGen is more stable, faster to train, and rated higher subjectively despite being more repetitive.

Load-bearing premise

The central claim rests on the assumption that the FAD and FD scores can reliably rank adapter configurations: each configuration is scored once against a fixed 400-clip reference set with no error bars, repeated seeds, or statistical significance test, so the few-point gaps that separate the "optimal" 40M configuration from its neighbors could be measurement noise.

Editorial extensions

If this is right

  • A practitioner can adapt MusicGen to a new genre in about 20-22 GPU hours and Mustango in about 45 GPU hours at the 40M scale, instead of full fine-tuning over days.
  • The best configuration is genre-dependent: 40M CNN adapters for Hindustani Classical, 40M transformer adapters for Turkish Makam, and 40M CNN adapters for Mustango on both genres.
  • Middle-layer placement should be avoided entirely; placing adapters at block ends or late layers is a precondition for coherent generation.
  • Model choice involves a real trade-off: Mustango gives broader instrument, rhythm, and melody diversity with better prompt adherence, while MusicGen gives more stable, clearer, subjectively higher-quality audio with more repetition.
  • Linear adapters are the weakest option for MusicGen and are not usable with Mustango's continuous latent representation.

Reading between the lines

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

  • A direct test of the capacity story: repeat the size sweep with 2x and 4x more training data; if the FAD/FD minimum moves above 40M, the sweet spot is set by data volume rather than by model architecture.
  • The placement result suggests that early and middle layers of these models carry genre-agnostic core musical representations; probing layer activations before and after adaptation could identify which layers are actually load-bearing.
  • Because the 40M ranking rests on single FAD/FD values, bootstrapping the 400-prompt reference set or repeating each configuration with several seeds would reveal whether the sweet spot is a real valley or measurement noise.
  • Feeding Mustango the chord and beat conditioning it was designed for, alongside the text-only prompts used here, may close its stability gap with MusicGen while keeping its diversity advantage.
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

5 major / 5 minor

Summary. The paper studies adapter-based parameter-efficient fine-tuning (PEFT) of two music generation models, MusicGen (autoregressive transformer) and Mustango (latent diffusion), for two low-resource non-Western genres, Hindustani Classical and Turkish Makam. It compares three adapter architectures (linear/MLP, CNN, transformer) at parameter scales from roughly 2M to 70M, and two placement strategies (middle vs. late layers). The evaluation uses Fréchet Audio Distance (FAD) and Fréchet Distance (FD) computed once per configuration on a fixed 400-clip prompt corpus, a two-annotator arena-style subjective evaluation, and GPU-hour measurements. The central claims are that a 40M-parameter adapter scale offers the best quality/efficiency trade-off, that late-layer placement preserves generation quality while middle-layer placement breaks it, that CNN adapters suit Hindustani Classical while transformer adapters suit Turkish Makam, and that MusicGen is faster to train and produces cleaner audio while Mustango is more diverse and prompt-adherent but less stable.

Significance. If the empirical findings are reliable, the paper provides practical, cost-sensitive guidance for adapting large music models to underrepresented genres: roughly 20-45 GPU hours instead of full fine-tuning, with concrete architectural and placement recommendations. The release of datasets, models, and training code is a concrete reproducibility asset. However, the headline '40M sweet spot' and the architecture/placement conclusions rest on an evaluation instrument whose precision is not demonstrated: each FAD/FD value is a single measurement with no error bars, the subjective evaluation uses two annotators without agreement statistics, and middle-layer failures are reported without quantitative support. These gaps directly affect the load-bearing claims, so the paper's contribution is currently conditional on additional evidence rather than established.

major comments (5)
  1. [Section 3.3.1 and Figure 2] The central '40M sweet spot' claim in Section 4.1 is based on a single FAD/FD value per configuration, computed on a fixed 400-clip prompt corpus with no repeated seeds, no bootstrap over prompt subsets, and no significance test. The differences that select 40M are small in FAD units (e.g., Hindustani_MG-C at 10.0, Makam_MG-T at 13.0, Mustango Hindustani MS-C at 6.4), and neighboring values are not shown, so it is impossible to rule out run-to-run or reference-set noise as the source of the ranking. Please provide confidence intervals or repeated-seed results, and report the full table of FAD/FD values across all sizes and architectures. Additionally, the metric is computed with PANN-CNN14 features through the AudioLDM framework rather than the canonical VGGish-based FAD; please justify this choice or report both versions, since it changes the metric's scale and sensitivity.
  2. [Sections 2.4.1 and 2.4.2] The placement contribution — late layers preserve quality, middle layers break generation — is reported only qualitatively for both MusicGen and Mustango. The text states that middle-layer placements led to 'complete breakdown' and 'significantly higher' FAD/FD values, but no numbers, table, or figure are provided for these failed configurations. This is load-bearing because the late-layer placement recommendation is a primary contribution. Please include the quantitative FAD/FD and/or listening outcomes for at least the middle-layer versus late-layer comparisons, with the same statistical treatment recommended above.
  3. [Section 3.3.2 and Figure 3] The subjective evaluation uses two annotators with no reported inter-annotator agreement, no per-annotator breakdown, and only 10 prompts per genre. The paper itself flags the annotator count as a resource constraint, but the resulting rankings (MG-C best for Hindustani, MG-T best for Turkish Makam) are then used as evidence for architecture conclusions. Please report an agreement statistic (e.g., Cohen's kappa) and the per-prompt or per-annotator results; without this, the subjective ranking is not distinguishable from chance-level variability.
  4. [Section 4.1 and Section 3.1] The '40M sweet spot' is presented as an optimal trade-off 'for the given amount of data', but the experimental design does not vary the dataset size: each genre has one fixed training set size (208.58 hours for Hindustani, 157.01 hours for Turkish Makam). The claim that 40M is optimal for these data amounts is therefore an observation about two points, not a scaling-law statement. To support the general interpretation, the paper should either vary the training data size across at least two or three amounts or explicitly limit the claim to the specific dataset sizes studied and avoid the implication of a broader scaling relationship.
  5. [Section 4.2] The conjecture that Mustango's instability stems from adapting it without chord or beat conditioning is plausible but untested. It is presented as a hypothesis in the discussion, yet the conclusion section restates it as a finding. Please either test this hypothesis (e.g., by conditioning Mustango with chord and beat metadata during fine-tuning) or clearly mark it as speculation; as written, it risks being read as an empirical result.
minor comments (5)
  1. [Figure 2] The caption states 'across three adapter architectures at varying parameter scales' but does not describe what the colors, markers, or line styles represent; a legend or explicit caption should be added so the reader can map each curve to a specific adapter type and genre.
  2. [Section 4.3] The sentence 'Mustango adapters (denoted MS- in the figure) require between 48 and 71 hours for MG-C and MG-T respectively' appears to contain a typo: it should refer to MS-C and MS-T, not MG-C and MG-T. Please also define the model names in the Figure 4 caption.
  3. [Conclusion] The conclusion states 'Mustango showed high FAD scores' for Turkish Makam, but Section 4.1 reports Makam_MS-C FAD as 8.39, which is numerically low; please reconcile these statements or clarify whether the comparison is relative to MusicGen or to the Hindustani values.
  4. [Section 2.3] For Mustango, the text says linear adapters are not used because 'MLPs require flattened 2D inputs', but a linear adapter could be applied pointwise to the channel dimension; this justification is not fully clear and would benefit from a one-sentence explanation of why pointwise MLPs are incompatible with the UNet's 3D latents.
  5. [References] Reference [13] is titled 'Fréchet Audio Distance: A Reference-Free Metric for Evaluating Music Enhancement Algorithms'; FAD is computed against a reference distribution, so the word 'reference-free' in the title is potentially confusing, but the citation itself appears accurate. Please verify the exact title.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the paper is an empirical adapter comparison whose claims rest on external metrics and new experiments, not on self-referential derivation.

full rationale

This paper makes no derivation or prediction claim that reduces to its own inputs. Its central assertions—the 40M parameter 'sweet spot,' the genre-dependent architecture preference (CNN for Hindustani, Transformer for Turkish Makam), and the placement findings—are empirical observations from FAD/FD measurements and human evaluations computed on test data that was held out from training. FAD and FD are external, pre-existing metrics evaluated with PANN-CNN14 features; the 40M optimum is read off the measured curves, not fitted or renamed from an input quantity. The self-citations to Mehta et al. [20] and [21] supply motivation (underrepresentation of the Global South) and an evaluation framework (arena-style subjective comparison), but neither import a conclusion: the subjective matchups are newly conducted with new annotations on new outputs, and the motivation claim is contextual rather than load-bearing for the quantitative trade-off findings. The lack of error bars or significance tests on the FAD/FD values is a legitimate internal-validity concern—the ranking could be noise—but that is a measurement-robustness issue, not a circularity issue, because the reported values are not derived from the conclusions they support. No equation in the paper equates an output to an input, no fitted parameter is renamed as a prediction, and no uniqueness or ansatz is imported from same-author prior work. The derivation chain, such as it is, is self-contained empirical measurement.

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

The paper is an empirical benchmarking study, so its ledger is dominated by measurement and sampling assumptions rather than fitted parameters. The free parameters are hand-chosen experimental settings (adapter bottleneck scales from 2M to 70M, learning rates, weight decays, epoch counts). The axioms are domain assumptions about metric reliability, model representativeness, dataset metadata quality, the post-hoc explanation for why middle-layer placement fails, and the sufficiency of two annotators. The paper introduces no new entities such as models, forces, or conserved quantities.

free parameters (4)
  • Adapter bottleneck dimension (per architecture and scale) = 2M to 70M trainable parameters; 40M highlighted as optimal
    Chosen by hand to sample a size range; the '40M sweet spot' claim depends on this discrete sampling and on the scale labels.
  • Learning rate = 5e-5 (MusicGen), 4.5e-5 (Mustango)
    Hand-picked hyperparameters reported in Section 3.2, not tuned per configuration.
  • Weight decay = 0.05 (MusicGen), 0.0001 (Mustango)
    Hand-picked; it affects the overfitting behavior that the size-optimality claims invoke.
  • Number of training epochs = 10-15 (Mustango), 18-25 (MusicGen)
    Set by early stopping on the validation set, with different ranges per model, which complicates cross-model comparisons.
assumptions (5)
  • domain assumption FAD and FD computed with PANN-CNN14 features on a fixed 400-clip prompt corpus reliably rank generation quality across configurations
    Stated in Section 3.3.1; there are no error bars, no repeated seeds, and no significance tests, so metric noise could exceed the small FAD differences used to pick winners.
  • domain assumption MusicGen and Mustango are representative of autoregressive and diffusion text-to-music model families
    Section 2.1 selects one model per family, yet the conclusions are framed per architecture type.
  • domain assumption Dunya corpus metadata (raga, makam, taal, usul, instruments) is accurate and sufficient for forming conditioning prompts
    Section 3.1; prompt fidelity is load-bearing for the genre-expression and prompt-adherence claims.
  • ad hoc to paper Middle layers encode foundational musical concepts whose modification disrupts the model
    Section 2.4.2; this post-hoc hypothesis explains the placement failures but no activation-level evidence is provided.
  • domain assumption Two annotators are sufficient for subjective quality ranking
    Section 3.3.2; the paper self-flags the constraint and asserts sufficiency without agreement statistics.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Exploring Adapter Design Tradeoffs for Low Resource Music Generation." pith.science (2026). https://pith.science/paper/GJOG44Y5

@misc{pith2026250621298,
  author       = {Pith},
  title        = {Pith review of: Exploring Adapter Design Tradeoffs for Low Resource Music Generation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GJOG44Y5}},
  note         = {Machine review of arXiv:2506.21298}
}
read the original abstract

Fine-tuning large-scale music generation models, such as MusicGen and Mustango, is a computationally expensive process, often requiring updates to billions of parameters and, therefore, significant hardware resources. Parameter-Efficient Fine-Tuning (PEFT) techniques, particularly adapter-based methods, have emerged as a promising alternative, enabling adaptation with minimal trainable parameters while preserving model performance. However, the design choices for adapters, including their architecture, placement, and size, are numerous, and it is unclear which of these combinations would produce optimal adapters and why, for a given case of low-resource music genre. In this paper, we attempt to answer this question by studying various adapter configurations for two AI music models, MusicGen and Mustango, on two genres: Hindustani Classical and Turkish Makam music. Our findings reveal distinct trade-offs: convolution-based adapters excel in capturing fine-grained local musical details such as ornamentations and short melodic phrases, while transformer-based adapters better preserve long-range dependencies crucial for structured improvisation. Additionally, we analyze computational resource requirements across different adapter scales, demonstrating how mid-sized adapters (40M parameters) achieve an optimal balance between expressivity and quality. Furthermore, we find that Mustango, a diffusion-based model, generates more diverse outputs with better adherence to the description in the input prompt while lacking in providing stability in notes, rhythm alignment, and aesthetics. Also, it is computationally intensive and requires significantly more time to train. In contrast, autoregressive models like MusicGen offer faster training and are more efficient, and can produce better quality output in comparison, but have slightly higher redundancy in their generations.

Figures

Figures reproduced from arXiv: 2506.21298 by the authors.

Figure 1
Figure 1. Adapter-based fine-tuning: Exploring different placements and architecture types (fully connected layers, Convolution, [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Comparing FAD and FD scores for MusicGen & Mustango across three adapter architectures at varying parameter scales for Hindustani Classical and Turkish Makam music. To maintain uniformity and optimize computational performance, we processed the dataset by shortening longer recordings into 30- second segments while preserving all metadata. These metadata elements, rich in genre-specific characteristics, were embedded… view at source ↗
Figure 3
Figure 3. Human evaluation of subjective quality and aesthetics for (a) Hindustani Classical and (b) Turkish Makam music. [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: GPU hours used across five adapter models at dif [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

38 extracted references · 21 canonical work pages

  1. [1]

    Andrea Agostinelli, Timo I Denk, Zalán Borsos, Jesse Engel, Mauro Verzetti, An- toine Caillon, Qingqing Huang, Aren Jansen, Adam Roberts, Marco Tagliasacchi, et al. 2023. Musiclm: Generating music from text. arXiv preprint arXiv:2301.11325 (2023)

    arXiv 2023

  2. [2]

    Negar Arabzadeh and Charles Clarke. 2024. Fréchet Distance for Offline Eval- uation of Information Retrieval Systems with Sparse Labels. In Proceedings of the 18th Conference of the European Chapter of the Association for Computa- tional Linguistics (Volume 1: Long Papers) , Yvette Graham and Matthew Purver (Eds.). Association for Computational Linguistic...

    work page 2024

  3. [3]

    Jean-Julien Aucouturier and Francois Pachet. 2003. Representing musical genre: A state of the art. Journal of new music research 32, 1 (2003), 83–93

    work page 2003

  4. [4]

    Judith Becker. 1993. Gamelan Stories: Tantrism, Islam, and Aesthetics in Central Java. (1993)

    work page 1993

  5. [5]

    Jade Copet, Felix Kreuk, Itai Gat, Tal Remez, David Kant, Gabriel Synnaeve, Yossi Adi, and Alexandre Defossez. 2023. Simple and Controllable Music Genera- tion. In Advances in Neural Information Processing Systems , A. Oh, T. Naumann, A. Globerson, K. Saenko, M. Hardt, and S. Levine (Eds.), Vol. 36. Curran Asso- ciates, Inc., 47704–47720. https://proceedi...

    work page 2023

  6. [6]

    Dan Hendrycks and Kevin Gimpel. 2016. Gaussian error linear units (gelus). arXiv preprint arXiv:1606.08415 (2016)

    arXiv 2016

  7. [7]

    Jordan Hoffmann, Sebastian Borgeaud, Arthur Mensch, and et al. 2022. Train- ing Compute-Optimal Large Language Models. arXiv preprint arXiv:2203.15556 (2022)

    arXiv 2022

  8. [8]

    Neil Houlsby, Andrei Giurgiu, Stanislaw Jastrzebski, Bruna Morrone, Quentin De Laroussilhe, Andrea Gesmundo, Mona Attariyan, and Sylvain Gelly. 2019. Parameter-efficient transfer learning for NLP. In International conference on machine learning. PMLR, 2790–2799

    2019

Show all 38 references
  1. [9]

    Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, Weizhu Chen, et al. 2022. Lora: Low-rank adaptation of large language models. ICLR 1, 2 (2022), 3

    2022

  2. [10]

    Jie Hu, Li Shen, and Gang Sun. 2018. Squeeze-and-excitation networks. InProceed- ings of the IEEE conference on computer vision and pattern recognition . 7132–7141

    2018

  3. [11]

    Jairazbhoy

    N.A. Jairazbhoy. 1971. The R ¯ags of North Indian Music: Their Structure and Evolution. Wesleyan University Press. https://books.google.ae/books?id= 0A0wAQAAIAAJ

    1971

  4. [12]

    Jared Kaplan, Sam McCandlish, Tom Henighan, and et al. 2020. Scaling Laws for Neural Language Models. arXiv preprint arXiv:2001.08361 (2020)

    2020 arXiv

  5. [13]

    Kevin Kilgour, Mauricio Zuluaga, Dominik Roblek, and Matthew Sharifi. 2019. Fréchet Audio Distance: A Reference-Free Metric for Evaluating Music Enhance- ment Algorithms. In Interspeech. https://api.semanticscholar.org/CorpusID: 202725406

    2019

  6. [14]

    Plumbley

    Qiuqiang Kong, Yin Cao, Turab Iqbal, Yuxuan Wang, Wenwu Wang, and Mark D. Plumbley. 2020. PANNs: Large-Scale Pretrained Audio Neural Networks for Audio Pattern Recognition. IEEE Transactions on Audio, Speech, and Language Processing 28 (Oct. 2020), 2880–2894. doi:10.1109/TASLP...

    2020

  7. [15]

    Yann LeCun and Yoshua Bengio. 1998. Convolutional networks for images, speech, and time series. MIT Press, Cambridge, MA, USA, 255–258

    1998

  8. [16]

    Xiang Lisa Li and Percy Liang. 2021. Prefix-tuning: Optimizing continuous prompts for generation. arXiv preprint arXiv:2101.00190 (2021)

    2021 arXiv

  9. [17]

    Haohe Liu, Zehua Chen, Yi Yuan, Xinhao Mei, Xubo Liu, Danilo Mandic, Wenwu Wang, and Mark D Plumbley. 2023. AudioLDM: Text-to-Audio Generation with Latent Diffusion Models. Proceedings of the International Conference on Machine Learning (2023), 21450–21474

    2023

  10. [18]

    Zequan Liu, Jiawen Lyn, Wei Zhu, Xing Tian, and Yvette Graham. 2024. ALoRA: Allocating Low-Rank Adaptation for Fine-tuning Large Language Models. In Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Languag...

    2024 doi

  11. [19]

    Ilya Loshchilov and Frank Hutter. 2017. Decoupled Weight Decay Regular- ization. In International Conference on Learning Representations . https://api. semanticscholar.org/CorpusID:53592270

    2017

  12. [20]

    Atharva Mehta, Shivam Chauhan, and Monojit Choudhury. 2024. Missing Melodies: AI Music Generation and its "Nearly" Complete Omission of the Global South. arXiv:2412.04100 [cs.SD] https://arxiv.org/abs/2412.04100

    2024 arXiv

  13. [21]

    Atharva Mehta, Shivam Chauhan, Amirbek Djanibekov, Atharva Kulkarni, Gus Xia, and Monojit Choudhury. 2025. Music for All: Exploring Multicultural Repre- sentations in Music Generation Models. arXiv:2502.07328 [cs.SD]

    2025 arXiv

  14. [22]

    Jan Melechovsky, Zixun Guo, Deepanway Ghosal, Navonil Majumder, Dorien Herremans, and Soujanya Poria. 2024. Mustango: Toward Controllable Text-to- Music Generation. In Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguis...

    2024

  15. [23]

    Bruno Nettl. 2001. Music of the middle east. Excursions in world music (2001), 46–73

    2001

  16. [24]

    Jonas Pfeiffer, Aishwarya Kamath, Andreas Rücklé, Kyunghyun Cho, and Iryna Gurevych. 2020. Adapterfusion: Non-destructive task composition for transfer learning. arXiv preprint arXiv:2005.00247 (2020)

    2020 arXiv

  17. [25]

    Jonas Pfeiffer, Andreas Rücklé, Clifton Poth, Aishwarya Kamath, Ivan Vulić, Sebastian Ruder, Kyunghyun Cho, and Iryna Gurevych. 2020. Adapterhub: A framework for adapting transformers. arXiv preprint arXiv:2007.07779 (2020)

    2020 arXiv

  18. [26]

    Alastair Porter, Mohamed Sordo, and Xavier Serra. 2013. Dunya: a system for browsing audio music collections exploiting cultural context. http://hdl.handle. net/10230/32251

    2013

  19. [27]

    Lutz Prechelt. 2002. Early stopping-but when? In Neural Networks: Tricks of the trade. Springer, 55–69

    2002

  20. [28]

    Flavio Schneider, Ojasv Kamal, Zhijing Jin, and Bernhard Schölkopf. 2024. Moûsai: Efficient text-to-music diffusion models. InProceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) . 8050– 8068

    2024

  21. [29]

    Xavier Serra. 2014. Creating research corpora for the computational study of music: the case of the Compmusic project. In AES 53rd International Conference: Semantic Audio; 2014 Jan 27-29; London, UK. New York: Audio Engineering Society

    2014

  22. [30]

    K.L. Signell. 2008. Makam: Modal Practice in Turkish Art Music . Usul Editions. https://books.google.ae/books?id=-G5MPgAACAAJ

    2008

  23. [31]

    Or Tal, Alon Ziv, Itai Gat, Felix Kreuk, and Yossi Adi. 2024. Joint Audio and Symbolic Conditioning for Temporally Controlled Text-to-Music Generation. arXiv:2406.10970 [cs.SD] https://arxiv.org/abs/2406.10970

    2024 arXiv

  24. [32]

    Yan Tao, Olga Viberg, Ryan S Baker, and René F Kizilcec. 2024. Cul- tural bias and cultural alignment of large language models. PNAS Nexus 3, 9 (09 2024), pgae346. arXiv:https://academic.oup.com/pnasnexus/article- pdf/3/9/pgae346/59151559/pgae346.pdf doi:10.1093/pnasnexus/pgae346

    2024 doi

  25. [33]

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Ł ukasz Kaiser, and Illia Polosukhin. 2017. Attention is All you Need. In Advances in Neural Information Processing Systems , I. Guyon, U. Von Luxburg, S. Bengio, H. Wallach, R. Fergus, S. ...

    2017

  26. [34]

    KG Vijayakrishnan. 2007. The grammar of Carnatic music . Mouton de Gruyter

    2007

  27. [35]

    Juncheng Yang, Zuchao Li, Shuai Xie, Weiping Zhu, Wei Yu, and Shijun Li. 2024. Cross-modal adapter: Parameter-efficient transfer learning approach for vision- language models. In 2024 IEEE International Conference on Multimedia and Expo (ICME). IEEE, 1–6

    2024

  28. [36]

    Fisher Yu, Vladlen Koltun, and Thomas Funkhouser. 2017. Dilated residual networks. In Proceedings of the IEEE conference on computer vision and pattern recognition. 472–480

    2017

  29. [2014]

    Audio Engineering Society

    Article number 1-1 [9 p.]. Audio Engineering Society

  30. [8316]

    doi:10.18653/v1/2024.naacl-long.459

    2024 doi

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.