The reviewed record of science sign in
Pith

arxiv: 2606.00629 · v1 · pith:3RNH7XQN · submitted 2026-05-30 · cs.SD · cs.HC· cs.LG· eess.AS

Quality Audio Prototyping: a prototype system for unified sound retrieval and procedural generation

Nelly Garcia , Aditya Bhattacharjee , Gabryel Mason-Williams , Israel Mason-Williams , Emmanouil Benetos , Joshua Reiss This is my paper

Reviewed by Pith T0 review T1 audit T2 compute T3 formal T4 kernel 2026-06-28 18:14 UTCgrok-4.3pith:3RNH7XQNrecord.jsonopen to challenge →

classification cs.SD cs.HCcs.LGeess.AS
keywords audio retrievalprocedural generationsound designprototypeparameter guidanceuser evaluationsimilarity search
0
0 comments X

The pith

QuAP unifies audio library search with procedural synthesis in one interface to cut the gap from idea to sound.

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

The paper presents Quality Audio Prototyping, or QuAP, as a prototype system that combines similarity-based retrieval of audio content with real-time procedural generation models. It adds a rule-based assistant offering parameter recommendations based on empirical data to help users without requiring expert knowledge in synthesis. This setup is meant to make sound design more efficient by keeping everything in one place rather than switching between separate tools. Practitioners would care if it truly lowers the effort needed to turn a story concept into high-quality audio while keeping their creative control intact. Evaluations suggest the system improves sound quality and is useful in practice for those working with sound effects.

Core claim

QuAP integrates a similarity-based retrieval engine with real-time procedural audio models, complemented by a rule-based assistant that provides perceptually informed parameter guidance derived from empirical optimisation. This reduces the procedural distance between a narrative concept and its sonic realisation, as confirmed by subjective assessments showing quality improvements and user evaluations indicating preserved creative agency.

What carries the argument

The rule-based assistant that supplies perceptually informed parameter guidance derived from empirical optimisation, which offers definitions and recommendations to users.

If this is right

  • Quality improvements in five of six synthesis models were statistically significant.
  • The preferred retrieval architecture was identified via encoder ablation on a sound effect dataset.
  • All 16 practitioners in the user study found the tool useful for their workflow.
  • The parameter assistant lowered the barrier to procedural interaction without removing creative agency.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The system could be tested in professional production pipelines to measure time savings in real projects.
  • Extending the assistant to learn from individual user preferences might enhance its utility over time.
  • Similar unification approaches might apply to other creative domains like visual effects or music arrangement.

Load-bearing premise

The recommendations from the rule-based assistant are perceptually valid and maintain creative agency for different users and situations.

What would settle it

A study where users perform the same sound design task with and without the assistant, checking if the recommended parameters lead to sounds rated as higher quality by listeners or if users feel their creativity is restricted.

Figures

Figures reproduced from arXiv: 2606.00629 by Aditya Bhattacharjee, Emmanouil Benetos, Gabryel Mason-Williams, Israel Mason-Williams, Joshua Reiss, Nelly Garcia.

Figure 1
Figure 1. Figure 1: QuAP system architecture. The loading library stage (top left) processes the user’s audio library through an offline deep￾embedding extractor and FAISS indexing backend, executed asynchronously in a background thread. The real-time query stage (bottom left) accepts drag-and-drop or text input, performing similarity search against the indexed database. Where procedural models are available, the system route… view at source ↗
Figure 2
Figure 2. Figure 2: Feature-driven bottleneck framework used to optimise procedural audio model parameters. Steps 1–3 correspond to one-versus￾all classification, top-K feature importance regression, and top-K feature classification respectively [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: QuAP graphical user interface. The left panel displays the audio library search results, supporting both text-based and drag-and-drop similarity queries. The right panel exposes the pro￾cedural audio model controls, with the embedded assistant provid￾ing perceptually informed parameter recommendations [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
read the original abstract

Sound design workflows frequently oscillate between time-consuming library searches and the complexity of procedural synthesis, with practitioners typically relying on disconnected tools to address each challenge separately. This paper introduces Quality Audio Prototyping (QuAP), a working prototype that unifies content-based audio retrieval and procedural sound generation within a single interface, reducing the procedural distance between a narrative concept and its sonic realisation. QuAP integrates a similarity-based retrieval engine with real-time procedural audio models, complemented by a rule-based assistant that provides perceptually informed parameter guidance, offering definitions and recommendations derived from empirical optimisation rather than requiring prior synthesis knowledge. Preliminary evaluation confirms the viability of this approach: subjective assessment demonstrated statistically significant quality improvements in five of six embedded synthesis models, and an encoder ablation study established the preferred retrieval architecture on a sound effect dataset. A user evaluation with 16 practitioners confirmed the tool's workflow utility, with all participants agreeing that the parameter assistant preserved creative agency while lowering the barrier to procedural interaction.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The paper presents Quality Audio Prototyping (QuAP), a working prototype that integrates similarity-based audio retrieval, real-time procedural sound generation models, and a rule-based assistant providing perceptually informed parameter guidance derived from empirical optimisation. It claims this unified interface reduces the procedural distance between narrative concepts and sonic realisations. Preliminary evaluations include subjective assessments reporting statistically significant quality improvements in five of six synthesis models, an encoder ablation study on a sound effect dataset, and a user study with 16 practitioners showing unanimous agreement that the assistant preserves creative agency while lowering interaction barriers.

Significance. If the empirical results hold after full methodological disclosure, the work offers a practical systems contribution to sound design by bridging disconnected retrieval and synthesis workflows. The combination of content-based retrieval with rule-guided procedural models could lower barriers for practitioners, and the reported user agreement on creative agency preservation would support broader adoption of such prototypes. The ablation study on retrieval architectures provides a concrete, if preliminary, comparison point for future systems.

major comments (3)
  1. [Abstract and Evaluation] Abstract and Evaluation section: The claim of statistically significant quality improvements in five of six embedded synthesis models is presented without methods detail, error bars, dataset sizes, participant numbers, exclusion criteria, or the specific statistical tests employed. This information is load-bearing for the central viability claim and must be supplied to allow assessment of reproducibility.
  2. [Rule-based assistant] Rule-based assistant description: The assistant is described as supplying recommendations 'derived from empirical optimisation' that are 'perceptually informed,' yet the manuscript provides no information on the optimisation dataset, objective function, participant pool, statistical controls, or the exact mapping from optimisation outputs to rules. This directly underpins the claim of perceptual validity and preservation of creative agency.
  3. [User evaluation] User evaluation: The report of unanimous agreement among 16 practitioners that the parameter assistant preserved creative agency lacks any description of the questionnaire design, response format, bias controls, or analysis procedure. Without these, the user-study results cannot be evaluated or replicated.
minor comments (2)
  1. [Introduction] The term 'procedural distance' is introduced in the abstract and introduction but is not formally defined or operationalised in the text, which may confuse readers unfamiliar with the concept.
  2. [Ablation study] Figure captions and axis labels in the ablation study results should explicitly state the dataset size and the exact metric used for 'preferred retrieval architecture' to improve clarity.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the detailed and constructive report. We agree that the current manuscript requires expanded methodological disclosure to support the evaluation claims and will revise accordingly.

read point-by-point responses

  1. Referee: [Abstract and Evaluation] The claim of statistically significant quality improvements in five of six embedded synthesis models is presented without methods detail, error bars, dataset sizes, participant numbers, exclusion criteria, or the specific statistical tests employed. This information is load-bearing for the central viability claim and must be supplied to allow assessment of reproducibility.

    Authors: We acknowledge that the evaluation section in the current draft does not provide sufficient methodological detail. In the revised manuscript we will add a dedicated subsection under Evaluation that reports participant numbers, exclusion criteria, dataset sizes, error bars on all relevant figures, and the exact statistical tests (including p-values) used to establish significance in five of the six models. This will directly address reproducibility concerns. revision: yes

  2. Referee: [Rule-based assistant] The assistant is described as supplying recommendations 'derived from empirical optimisation' that are 'perceptually informed,' yet the manuscript provides no information on the optimisation dataset, objective function, participant pool, statistical controls, or the exact mapping from optimisation outputs to rules. This directly underpins the claim of perceptual validity and preservation of creative agency.

    Authors: We agree that the description of the rule-based assistant is currently underspecified. The revised manuscript will include a new subsection detailing the optimisation dataset, the objective function employed, the participant pool, statistical controls applied, and the precise mapping procedure from optimisation results to the implemented rules. This addition will substantiate the perceptual grounding of the assistant. revision: yes

  3. Referee: [User evaluation] The report of unanimous agreement among 16 practitioners that the parameter assistant preserved creative agency lacks any description of the questionnaire design, response format, bias controls, or analysis procedure. Without these, the user-study results cannot be evaluated or replicated.

    Authors: We concur that the user-study reporting is incomplete. In revision we will expand the User Evaluation section to describe the questionnaire design, response format (including scale type), bias controls (such as question randomisation), and the analysis procedure used to arrive at the reported agreement levels. The existing participant count of 16 will be retained with these additional methodological details. revision: yes

Circularity Check

0 steps flagged

No circularity: systems prototype with empirical evaluations only

full rationale

The paper presents a prototype system (QuAP) integrating retrieval, procedural models, and a rule-based assistant, with claims supported by subjective user assessments and an encoder ablation study on a sound effect dataset. No mathematical derivations, equations, fitted predictions, or self-citations appear in the provided text. The rule-based assistant is described as using recommendations 'derived from empirical optimisation,' but this is presented as an external process without any reduction of outputs to inputs by construction within the paper itself. The derivation chain is self-contained against external benchmarks (user studies), with no load-bearing steps that match the enumerated circularity patterns.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

No mathematical model or derivation is present; the paper is an engineering prototype description.

pith-pipeline@v0.9.1-grok · 5727 in / 1118 out tokens · 23824 ms · 2026-06-28T18:14:39.552715+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

42 extracted references · 5 canonical work pages · 2 internal anchors

  1. [1]

    INTRODUCTION Sound design encompasses the creation and manipulation of envi- ronmental sounds, layered textures, and sound elements that com- pose the auditory dimension of audiovisual productions, distinct from dialogue and music [1]. To create a soundscape, practitioners rely on extensive sound libraries, Foley [2] recordings, and pro- duction stems, na...

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  2. [2]

    In professional practice, these two axes remain largely disconnected

    RELATED WORK The evolution of sound design tools has developed along two pri- mary axes: the ease of retrieval from large-scale audio databases, and the flexibility of real-time sound synthesis. In professional practice, these two axes remain largely disconnected. Practitioners typically navigate between separate retrieval platforms, synthesis tools, gene...

  3. [3]

    METHODOLOGY The development of QuAP required the optimisation of two in- dependent but complementary components: the procedural audio models that underpin the synthesis engine, and the audio embed- ding model that drives similarity-based retrieval. For the procedu- ral models, optimisation was guided by a feature-driven bottleneck framework [21], which id...

  4. [4]

    An overview of the system architecture is presented in Figure 1

    SYSTEM DESIGN The system integrates four primary components: (1) an offline embedding and indexing pipeline, (2) a real-time query inference module, (3) an interface for interaction with procedural models, and (4) a hybrid layering stage enabling the user to combine re- trieved library samples with procedurally generated audio. An overview of the system a...

    2026

  5. [5]

    All participants provided informed consent prior to taking part, were informed of their right to withdraw at any time, and all responses were anonymised before analysis

    USER EV ALUATION Ethical approval for this study was obtained in accordance with in- stitutional guidelines. All participants provided informed consent prior to taking part, were informed of their right to withdraw at any time, and all responses were anonymised before analysis. QuAP was evaluated by 16 participants following up their in- terest in the sur...

  6. [6]

    Our primary encoder is based on MobileNetV3, motivated by the findings of Greif et al

    ABLATION STUDY The choice of audio embedding architecture is guided by two de- sign considerations: the model should be lightweight enough for efficient deployment, while maintaining strong performance on re- lated content-based audio retrieval tasks. Our primary encoder is based on MobileNetV3, motivated by the findings of Greif et al. [16], who demonstr...

  7. [7]

    To ensure a fair com- parison, the baseline model is trained using the same data prepro- cessing and training procedure as the MobileNet-based encoder described in Section 4.3

    and music sample identification [33]. To ensure a fair com- parison, the baseline model is trained using the same data prepro- cessing and training procedure as the MobileNet-based encoder described in Section 4.3. Both models are evaluated on a held-out test split of the FSD50K dataset. Retrieval performance is measured using mean average precision (mAP)...

  8. [8]

    gives the sensation of keeping human-in-the-loop

    DISCUSSION The results across the three evaluation components — procedu- ral, model optimisation, encoder ablation, and user evaluation — collectively support the core premise of QuAP: that unifying re- trieval and synthesis within a single, interpretable environment is both technically viable and practically valuable for sound design workflows. 8https://...

    2026

  9. [9]

    CONCLUSION This paper introduced Quality Audio Prototyping (QuAP), a work- ing prototype application that unifies content-based audio retrieval and procedural sound generation within a single environment. By combining a MobileNetV3-based similarity search engine with six optimised procedural audio models and an embedded parameter assistant, QuAP addresses...

  10. [10]

    Sonnenschein,Sound Design: The Expressive Power of Music, Voice, and Sound Effects in Cinema

    D. Sonnenschein,Sound Design: The Expressive Power of Music, Voice, and Sound Effects in Cinema. Studio City, CA: Michael Wiese Productions, 2001

    2001

  11. [11]

    Ament,The Foley Grail: The Art of Performing Sound for Film, Games, and Animation

    V . Ament,The Foley Grail: The Art of Performing Sound for Film, Games, and Animation. Taylor & Francis, 2009. [Online]. Available: https://books.google.co.uk/books?id= sbcpAAAAQBAJ

    2009

  12. [12]

    Sound designer-generative ai interactions: Towards designing creative support tools for professional sound designers,

    P. Kamath, F. Morreale, P. L. Bagaskara, Y . Wei, and S. Nanayakkara, “Sound designer-generative ai interactions: Towards designing creative support tools for professional sound designers,” inProceedings of the 2024 CHI Confer- ence on Human Factors in Computing Systems, ser. CHI ’24. New York, NY , USA: Association for Computing Machinery, 2024

    2024

  13. [13]

    Billion-scale similarity search with gpus,

    J. Johnson, M. Douze, and H. Jégou, “Billion-scale similarity search with gpus,”IEEE transactions on big data, vol. 7, no. 3, pp. 535–547, 2019

    2019

  14. [14]

    Musiclm: Gen- erating music from text,

    A. Agostinelli, T. Denk, Z. Borsos, J. Engel, M. Verzetti, A. Caillon, Q. Huang, A. Jansen, A. Roberts, M. Tagliasac- chi, M. Sharifi, N. Zeghidour, and C. Frank, “Musiclm: Gen- erating music from text,” 01 2023

    2023

  15. [15]

    Audioldm: Text-to-audio generation with la- tent diffusion models,

    H. Liuet al., “Audioldm: Text-to-audio generation with la- tent diffusion models,”Proceedings of the International Con- ference on Machine Learning (ICML), 2023

    2023

  16. [16]

    Neural audio synthesis for sound effects: A scope review,

    M. Cámara, F. Marcos, A. R. Bargum, C. Erkut, J. Reiss, and J. L. Blanco, “Neural audio synthesis for sound effects: A scope review,”IEEE Transactions on Audio, Speech and Language Processing, vol. 34, pp. 427–445, 2026

    2026

  17. [17]

    An industrial-strength audio search algorithm,

    A. Wang, “An industrial-strength audio search algorithm,” in ISMIR, 2003

    2003

  18. [18]

    Neural audio fingerprint for high-specific audio retrieval based on contrastive learning,

    S. Chang, D. Lee, J. Park, H. Lim, K. Lee, K. Ko, and Y . Han, “Neural audio fingerprint for high-specific audio retrieval based on contrastive learning,” inICASSP 2021-2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2021, pp. 3025–3029

    2021

  19. [19]

    Cross recurrence quantification for cover song identification,

    J. Serra, X. Serra, and R. G. Andrzejak, “Cross recurrence quantification for cover song identification,”New Journal of Physics, vol. 11, no. 9, p. 093017, 2009

    2009

  20. [20]

    Accurate and scal- able version identification using musically-motivated embed- dings,

    F. Yesiler, J. Serrà, and E. Gómez, “Accurate and scal- able version identification using musically-motivated embed- dings,” inICASSP 2020-2020 IEEE International Confer- ence on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2020, pp. 21–25

    2020

  21. [21]

    Sound retrieval from voice imi- tation queries in collaborative databases,

    D. S. Blancas and J. Janer, “Sound retrieval from voice imi- tation queries in collaborative databases,” inAudio Engineer- ing Society Conference: 53rd International Conference: Se- mantic Audio. Audio Engineering Society, 2014

    2014

  22. [22]

    Imisound: An unsupervised system for sound query by vocal imitation,

    Y . Zhang and Z. Duan, “Imisound: An unsupervised system for sound query by vocal imitation,” in2016 IEEE Interna- tional Conference on Acoustics, Speech and Signal Process- ing (ICASSP). IEEE, 2016, pp. 2269–2273

    2016

  23. [23]

    Panns: Large-scale pretrained audio neural net- works for audio pattern recognition,

    Q. Kong, Y . Cao, T. Iqbal, Y . Wang, W. Wang, and M. D. Plumbley, “Panns: Large-scale pretrained audio neural net- works for audio pattern recognition,”IEEE/ACM Transac- tions on Audio, Speech, and Language Processing, vol. 28, pp. 2880–2894, 2020

    2020

  24. [24]

    MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications

    A. G. Howard, M. Zhu, B. Chen, D. Kalenichenko, W. Wang, T. Weyand, M. Andreetto, and H. Adam, “Mobilenets: Effi- cient convolutional neural networks for mobile vision appli- cations,”arXiv preprint arXiv:1704.04861, 2017

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  25. [25]

    Improving query-by-vocal imitation with contrastive learning and audio pretraining,

    J. Greif, F. Schmid, P. Primus, and G. Widmer, “Improving query-by-vocal imitation with contrastive learning and audio pretraining,”arXiv preprint arXiv:2408.11638, 2024

    work page arXiv 2024

  26. [26]

    Querying by vocal imitation chal- lenge,

    N. Garcia-Sihuay, D. Martilla, Y . Zhong, M. Liu, J. Groff, B. Swanson, and J. Reiss, “Querying by vocal imitation chal- lenge,”Journal of the Audio Engineering Society, 2026, un- der review

    2026

  27. [27]

    Farnell,Designing Sound

    A. Farnell,Designing Sound. MIT Press, 2010. [On- line]. Available: https://books.google.co.uk/books?id= JsuMEAAAQBAJ

    2010

  28. [28]

    The state of the art in procedural audio,

    D. Menexopoulos, P. Pestana, and J. Reiss, “The state of the art in procedural audio,”Journal of the Audio Engineering Society, vol. 71, pp. 825–847, 12 2023

    2023

  29. [29]

    Musicians’ ethical con- cerns about ai: an interview study,

    J. Herington, R. Borasi, B. Guerrero, D. Miller, B. Koerner, Y . J. Han, Z. Borys, and R. Roberts, “Musicians’ ethical con- cerns about ai: an interview study,”AI and SOCIETY, vol. 41, pp. 1075–1088, 09 2025

    2025

  30. [30]

    Concept Bottleneck Models

    P. W. Koh, T. Nguyen, Y . S. Tang, S. Mussmann, E. Pierson, B. Kim, and P. Liang, “Concept bottleneck models,” 2020. [Online]. Available: https://arxiv.org/abs/2007.04612 DAFx.7 Proceedings of the 29th International Conference on Digital Audio Effects (DAFx26), Cambridge, MA, USA, 1–4 September 2026

    work page Pith review arXiv 2020

  31. [31]

    Essentia: an open-source library for sound and music anal- ysis,

    D. Bogdanov, N. Wack, E. Gómez, S. Gulati, P. Herrera, O. Mayor, G. Roma, J. Salamon, J. Zapata, and X. Serra, “Essentia: an open-source library for sound and music anal- ysis,” inProceedings of the 21st ACM International Confer- ence on Multimedia. New York, NY , USA: Association for Computing Machinery, 2013, p. 855–858

    2013

  32. [32]

    6ksfx synth dataset,

    N. Garcia and J. Reiss, “6ksfx synth dataset,” no. arXiv:2501.17198, 2025

    work page arXiv 2025

  33. [33]

    Statistical tests with mushra data,

    C. Mendonça and S. Delikaris-Manias, “Statistical tests with mushra data,” 05 2018

    2018

  34. [34]

    Synthetic audio optimization by feature-driven importance,

    N. Garcia-Sihuay, G. Mason-Williams, I. Mason-Williams, and J. Reiss, “Synthetic audio optimization by feature-driven importance,”IEEE/ACM Transactions on Audio, Speech, and Language Processing, 2026, under review

    2026

  35. [35]

    Au- dio Set: An ontology and human-labeled dataset for audio events,

    J. F. Gemmeke, D. P. W. Ellis, D. Freedman, A. Jansen, W. Lawrence, R. C. Moore, M. Plakal, and M. Ritter, “Au- dio Set: An ontology and human-labeled dataset for audio events,” inProceedings of the IEEE International Confer- ence on Acoustics, Speech and Signal Processing (ICASSP), New Orleans, LA, 2017, pp. 776–780

    2017

  36. [36]

    Efficient large- scale audio tagging via transformer-to-cnn knowledge dis- tillation,

    F. Schmid, K. Koutini, and G. Widmer, “Efficient large- scale audio tagging via transformer-to-cnn knowledge dis- tillation,” inICASSP 2023-2023 IEEE international Confer- ence on acoustics, Speech and signal processing (ICASSP). IEEE, 2023, pp. 1–5

    2023

  37. [37]

    Supervised con- trastive learning,

    P. Khosla, P. Teterwak, C. Wang, A. Sarna, Y . Tian, P. Isola, A. Maschinot, C. Liu, and D. Krishnan, “Supervised con- trastive learning,”Advances in neural information process- ing systems, vol. 33, pp. 18 661–18 673, 2020

    2020

  38. [38]

    Fsd50k: an open dataset of human-labeled sound events,

    E. Fonseca, X. Favory, J. Pons, F. Font, and X. Serra, “Fsd50k: an open dataset of human-labeled sound events,” IEEE/ACM Transactions on Audio, Speech, and Language Processing, vol. 30, pp. 829–852, 2021

    2021

  39. [39]

    Qualitative HCI research: Going behind the scenes,

    A. Blandford, D. Furniss, and S. Makri, “Qualitative HCI research: Going behind the scenes,”Synthesis Lectures on Human-Centered Informatics, vol. 9, no. 1, pp. 1–147, 2016

    2016

  40. [40]

    Using thematic analysis in psychol- ogy,

    V . Braun and V . Clarke, “Using thematic analysis in psychol- ogy,”Qualitative research in psychology, vol. 3, no. 2, pp. 77–101, 2006

    2006

  41. [41]

    Bytecover: Cover song identification via multi-loss training,

    X. Du, Z. Yu, B. Zhu, X. Chen, and Z. Ma, “Bytecover: Cover song identification via multi-loss training,” inICASSP 2021-2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2021, pp. 551–555

    2021

  42. [42]

    Refining music sample identification with a self-supervised graph neural network,

    A. Bhattacharjee, I. Meresman Higgs, M. Sandler, and E. Benetos, “Refining music sample identification with a self-supervised graph neural network,” inProceedings of the 26th International Society for Music Information Retrieval Conference (ISMIR). ISMIR, 2025. DAFx.8

    2025