Pith. sign in

REVIEW 4 major objections 6 minor 44 references

Rethinking Automatic Music Mixing as Sequential Stem Blending

T0 review · 4 major / 6 minor · reviewed 2026-08-08 · deepseek-v4-flash

Pith's one-line read Sequential stem blending can replace single-pass music mixing: a flow-matching model integrates one stem at a time into a fixed submix, outperforming parallelized baselines on both stem blending and full automatic mixing benchmarks.

desk verdict The sequential stem blending formulation is a genuinely new idea worth taking seriously, but the paper's headline performance claim rests on an in-distribution benchmark built from the same degradation pipeline used for training. read the letter →

arxiv 2608.05506 v1 pith:VEV7T6GM submitted 2026-08-06 eess.AS

classification eess.AS
keywords automaticmusicmixingstemblendingflowmatchinglatentdiffusiondegradation-baseddatasynthesissequentialprocessingaudioeffectsstyle
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

Automatic music mixing has been treated as a one-shot operation: feed all stems into a network and get a mixture. This paper asks instead whether mixing can be decomposed into repeated single-stem blending steps, where each stem is fitted into a fixed submix, and answers yes. It trains a latent rectified flow-matching model on synthetic raw/wet blending pairs and shows competitive or better results on both stem blending and full mixing benchmarks. If the claim holds, mixing models become inspectable, interactive, and able to use stem ordering as a style control.

What carries the argument

The load-bearing mechanism is rectified flow matching in the latent space of a pretrained variational autoencoder, where the flow starts from the unprocessed stem latent $z_0$ instead of Gaussian noise and follows the interpolation $z_t = (1-t)z_0 + t z_1$ toward the processed stem latent, conditioned on the latent of the current submix plus genre, instrument, and loudness. The conditioning keeps the submix acoustically fixed, so each step integrates one stem into an anchor context. The paired training signal comes from a degradation-based synthesis that inverts common mixing decisions (masking, over-cutting, mud, harshness, blend, room reverb) to create realistic raw/wet/submix triplets.

What would settle it

Take a real session with true raw and wet stems plus the engineer's submix, and have an independent listening panel compare the model's processed stem against the true wet stem; if the model does no better than a model trained on real raw/wet pairs, the degradation synthesis is what limits full-mix quality. A complementary check: add real raw/wet pairs to training and see whether tonal-balance and style-similarity metrics on full mixes improve.

Watch

Extended reading notes

Core claim

The paper's central claim is that sequential stem blending is a principled and viable reformulation of automatic music mixing. A rectified flow matching model, conditioned on the current submix, transports the latent of an unprocessed stem to that of a processed stem while the submix stays fixed. Trained exclusively on degradation-synthesized pairs from MedleyDB and MoisesDB, the model achieves near-zero kernel audio distance on a stem-blending benchmark, outperforming parallelized baselines, and generalizes to full mixing where domain-knowledge ordering of stems improves coherence. The authors state explicitly that existing parallelized approaches are inherently not designed for stem blending.

Load-bearing premise

The whole method rests on the assumption that hand-crafted degradations applied to wet stems produce unprocessed stems similar enough to real ones that the model learns genuine mixing behavior rather than artifacts of the synthesis.

Editorial extensions

If this is right

  • A model trained only on single-stem blending can handle an arbitrary number of stems at inference by repeated application of the same blending step.
  • Users can blend a stem into their own submix, inspect intermediate submixes, or start the process from any point in the chain.
  • The order in which stems are processed changes the resulting mix, giving a training-free style control.
  • Parallelized baselines degrade when a stem is already well-suited to the mix, whereas sequential blending avoids this by anchoring on the fixed submix.
  • Domain-knowledge ordering (rhythm and foundation first) produces more coherent full mixtures than random ordering.

Reading between the lines

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

  • Because the model never sees sparse early submixes, a natural extension is a curriculum that trains on progressively emptier submixes; the paper itself notes this missing early-step exposure.
  • The degradation bank is hand-designed, so learned or automatically discovered degradations could push full-mix tonal balance closer to professional references.
  • The sequential formulation makes mixing a compositional, context-dependent process, which may transfer to interactive DAW tools and to style transfer between mixes by swapping submix contexts.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper proposes reformulating automatic music mixing (AMM) as sequential stem blending, in which each stem is transformed and added to a growing submix, inspired by professional mixing practice. The authors train a rectified latent flow matching model conditioned on the current submix, using a degradation-based data synthesis strategy to create training pairs from MedleyDB and MoisesDB. They evaluate on a new stem blending benchmark and on the standard full-AMM benchmark, reporting distributional metrics (KAD, FD) and a small listening test. The main claims are that parallelized AMM approaches are inherently unsuited to stem blending, that the proposed model achieves strong stem blending performance, and that it is competitive on full AMM while supporting interactive and inspectable workflows.

Significance. If the central claims were convincingly supported, the paper would make a useful contribution: the sequential formulation is intuitive, the latent flow matching model is technically sound, and the framing of stem ordering as a training-free style control is novel. The degradation-based data synthesis is a pragmatic way to create training targets, and the paper candidly acknowledges limitations in Section 7. However, the load-bearing quantitative evidence for stem blending superiority is produced by a benchmark generated with the same degradation pipeline used for training, making the main effectiveness claim currently unsupported outside that controlled setting. The paper's significance therefore depends on whether additional out-of-distribution evidence can be provided.

major comments (4)
  1. [§4.1 and §3.4] The stem blending benchmark is constructed using exactly the same degradation-based strategy described in Section 3.4 and applied to held-out MoisesDB stems during training. The near-zero KAD values in Table 1 therefore demonstrate that the model can invert or smooth the synthetic degradations it was trained on, not that it generalizes to real unprocessed stems and real mixing decisions. This benchmark is the sole quantitative support for the claim in Section 7 that 'our model achieves strong stem blending performance.' Please add an out-of-distribution objective evaluation, for example on MedleyDB raw/wet pairs with a clean train/test split, or on real raw stems from an external multitrack dataset; alternatively, re-frame the current stem blending results as an in-distribution sanity check rather than as evidence of practical mixing ability.
  2. [§6 and §4.4] The only out-of-distribution evidence is a MUSHRA-style listening test on three songs and six total samples with eighteen participants. Section 6 states that the proposed model 'achieves the highest median score in five of six examples' and that results are 'consistent,' but no significance testing is reported. Given the small sample size and the fact that this is the only non-circular evidence for generalization, please report statistical tests (e.g., Wilcoxon signed-rank with multiple-comparison correction) or explicitly characterize the results as anecdotal. Without such tests, the generalization claim in Section 6 ('the degradation-based training strategy captures generalizable mixing behavior') is not supported.
  3. [§5, Table 1] The proposed model's KAD values are negative (-0.01 for stem blending FxEnc++ and -0.07 for stem blending CLAP). KAD is based on MMD; if the estimator can be negative, this should be stated and confidence intervals or standard errors should be provided, because a negative distance is surprising and is not explained in the text. The near-zero magnitude, combined with the circular benchmark construction, reinforces the concern that the output and reference distributions are trivially close in this controlled setting rather than that the model is a strong blender.
  4. [§5 and §7] The conclusion that 'parallelized approaches are inherently not designed for this task' is too strong given the reported evidence. The re-blending variant (†) is a simple post-hoc adaptation in which only the predicted stem is summed with the original submix; this may disadvantage DMC and MEGAMI, but it does not establish that parallelized architectures are inherently incapable of stem blending. Furthermore, on the full AMM benchmark MEGAMI obtains better FxEncoder++ KAD and tonal balance FD (Table 1), and Section 7 concedes that full AMM performance 'does not yet match state-of-the-art systems' on those axes. Please temper the language and present the sequential approach as a promising complementary formulation with current limitations.
minor comments (6)
  1. [§3.6] The initialization of s(0) is described as a 'zero-valued latent vector' in the text but s(k) is defined as a waveform in Eq. (2); please clarify whether the initial submix is a zero waveform, a zero latent, or something else.
  2. [§4.3] Please provide more details on how KAD and the Fréchet distances are computed, including the number of samples used for each distribution, the embedding granularity, and whether the metrics are estimated on full 10-second segments or on shorter windows.
  3. [§6, Figure 3] Figure 3 shows MUSHRA score distributions but does not indicate the sample size per box or any statistical comparison; adding per-example sample sizes and significance brackets would make the figure more informative.
  4. [§2] In Eq. (3), the notation z_t = (1-t) z_0 + t z_1 appears before the loss is defined; consider labeling this as the interpolation formula to improve readability.
  5. [§3.4] The degradation modes are described qualitatively (e.g., 'masking boost,' 'harshness'); for reproducibility, please provide the exact parametric EQ settings, gain ranges, and room impulse response parameters used.
  6. [§3.6] The statement that training uses 'a batch size of 128' and '10-second audio segments at 44.1 kHz' on a single RTX 4090 is informative, but the total number of training steps or effective epochs should be reported for comparability with other flow-matching work.

Circularity Check

1 steps flagged · score 6.0 of 10

Stem-blending superiority rests on a benchmark built from the same hand-crafted degradation pipeline used for training; out-of-distribution evidence is a 6-sample listening test.

  1. fitted input called prediction [Section 3.4 and Section 4.1, Table 1]
    "Since the benchmark is constructed using the same degradation-based strategy described in Section 3.4, it provides a controlled setting to directly measure whether the model learns the intended blending transformation, with generalization to real-world scenarios assessed via perceptual evaluation on out-of-distribution stems (Section 4.4)."

    For MoisesDB, training pairs are synthesized by degrading wet stems: Section 3.4 states 'only wet stems are available, so we simulate x_k by applying degradations to y_k.' The held-out stem-blending benchmark applies the same degradation modes (masking boost, over-cut, low-end mud, harshness, blend, and room reverb) to wet stems to create the test 'unprocessed' stem, with the wet mix as reference. The model was trained to invert exactly this degradation distribution conditioned on the wet-stem submix, so the near-zero KAD and tiny FD values in Table 1 measure the model's ability to undo its own training augmentations on in-distribution inputs.

full rationale

The central circularity is that the paper's main evidence for stem-blending effectiveness is a benchmark generated by the same degradation-based synthesis used to create training data. In Section 3.4, MoisesDB training pairs are constructed by applying hand-crafted degradations to wet stems to simulate unprocessed stems; Section 4.1 then states that the stem blending benchmark is 'constructed using the same degradation-based strategy described in Section 3.4.' The model is therefore tested on the same transformation family it was trained to invert, and the near-perfect distributional scores in Table 1 are partly a measure of how well the model learned its own training augmentation. The paper is transparent about calling this a controlled setting, but the conclusion still relies on it for the load-bearing claim that sequential stem blending yields 'strong stem blending performance' and that parallelized baselines are 'inherently not designed' for the task. The non-circular evidence is weaker: the full-AMM benchmark on MedleyDB v2 is cleanly separated from training, but the results are only competitive and the paper concedes that full-AMM performance lags on tonal balance and mixing style similarity, with MEGAMI winning on FxEncoder++ KAD and tonal-balance FD. The perceptual test uses 3 songs, 6 samples, and 18 listeners with no significance testing, so it cannot independently support the strong claim. No load-bearing self-citation chain was found: FxEncoder++ is an author-contributed evaluation embedding, not an argument premise, and the flow-matching machinery is cited from independent prior work. Overall, the paper has independent content in its sequential formulation and MedleyDB v2 evaluation, but its flagship stem-blending success is partly circular by construction, giving a score of 6.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

No new physical entities, particles, or conserved quantities are introduced. The 'stem blending task' is a reformulation of an existing task, not an invented entity. Free parameters are the hand-designed degradation settings and inference hyperparameters that shape the training distribution and evaluation.

free parameters (3)
  • Degradation mode parameters (EQ bands/gains, reverb room sizes)
    Hand-chosen values for the five spectral degradation modes and room reverb define the training distribution; they are not fitted but are ad hoc choices that shape the target task. Section 3.4.
  • Inference steps = 10
    Number of ODE integration steps for rectified flow, chosen without a reported ablation. Section 3.6.
  • Classifier-free guidance drop probability = 0.1
    Dropout rate for submix condition masking; set by hand. Section 3.6.
assumptions (4)
  • domain assumption The Stable Audio Open VAE's latent space preserves acoustic structure while providing necessary stochasticity for flow matching.
    Relied on in Section 3.3; if the VAE latent is not sufficiently content-preserving and stochastic, the flow could either collapse to regression or corrupt musical content.
  • ad hoc to paper Degradation-based simulation of the final blending step is a sufficient training signal for real mixing transformations.
    Introduced in Section 3.4 and acknowledged in Section 7 as a likely cause of gaps in tonal balance and style similarity; this is the paper's own admission.
  • domain assumption Human mix engineers typically process stems one at a time, so sequential processing is a natural inductive bias.
    Motivational claim in Section 1; if false, the approach loses its conceptual grounding but not necessarily its empirical validity.
  • ad hoc to paper The instrument ordering (drums, bass, guitar, keys, strings, vocals, other) reflects a tonal and rhythmic dependency hierarchy.
    Hand-defined ordering in Section 3.5 used for Proposed-Domain inference.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Rethinking Automatic Music Mixing as Sequential Stem Blending." pith.science (2026). https://pith.science/paper/VEV7T6GM

@misc{pith2026260805506,
  author       = {Pith},
  title        = {Pith review of: Rethinking Automatic Music Mixing as Sequential Stem Blending},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VEV7T6GM}},
  note         = {Machine review of arXiv:2608.05506}
}
read the original abstract

Automatic music mixing, the task of automatically combining individual audio tracks into a cohesive mixture, is typically addressed by parallelized architectures that process all input tracks in a single pass. In this work, inspired by how human mix engineers process stems one at a time, we propose a paradigm shift and ask whether automatic music mixing can be reformulated as a sequential stem blending task, where each stem is blended into a growing submix. Specifically, we train a latent flow matching model conditioned on the submix context, enabling sequential processing of an arbitrary number of input tracks. To train the model, we introduce a degradation-based data synthesis strategy that simulates realistic stem blending scenarios from existing multitrack and source separation datasets. Experimental results on both stem blending and automatic music mixing benchmarks demonstrate the effectiveness of the proposed approach. We provide audio examples on the accompanying demo page\footnote{https://sequential-mixing-demo.vercel.app/}.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

44 extracted references · 34 canonical work pages

  1. [1]

    Rethinking Automatic Music Mixing as Sequential Stem Blending

    INTRODUCTION Music production is a complex creative process requiring domain expertise, particularly in the mixing stage, where individual audio tracks are combined through a series of audio effects into a cohesive mixture balancing frequency content, spatial image, dynamics, and loudness [1]. To make mixing more accessible, the field of automatic mu- sic...

  2. [2]

    PRELIMINARIES ON FLOW MA TCHING Flow matching defines a continuous mapping between a source distributionp 0 and a target distributionp 1 via an ordinary differential equation (ODE):dz t =v θ(zt, t)dt, wherev θ is a velocity field parameterized by a neural net- work with weightsθandt∈[0,1]is the continuous time step. We adopt rectified flow matching (RF) [...

  3. [3]

    METHOD 3.1 Problem Formulation Letx i ∈R 2×T denote thei-th unprocessed stem repre- sented as a stereo signal ofTsamples. Existing AMM systems adopt a parallelized paradigm that maps a set ofN such stemsX={x i}N i=1 to a final mixture ˆymix ∈R 2×T in a single pass: ˆymix =g {xi}N i=1 ,(1) where all stems are processed jointly in a single forward pass. In ...

  4. [4]

    EXPERIMENTAL SETUP 4.1 Evaluation Scenarios We evaluate on two benchmarks. For thestem blend- ing benchmark, a single stem is blended into the submix s(N−1) formed by the remaining stems, and the resulting mixture is evaluated against the professionally mixed ref- erence. Since the benchmark is constructed using the same degradation-based strategy describ...

  5. [5]

    Stem Blending.Table 1 presents the stem blending re- sults

    OBJECTIVE RESULTS While music mixing ultimately involves subjective judg- ment, the objective metrics in this section offer useful in- sights into the behavioral properties of each system, com- plementing the perceptual study in Section 6. Stem Blending.Table 1 presents the stem blending re- sults. Notably,Raw-mix outperforms all AMM base- lines despite a...

  6. [6]

    The proposed model achieves the highest score in both and the highest overall mean

    PERCEPTUAL EV ALUA TION Meta Audiobox Aesthetics.Table 2 presents the produc- tion quality (PQ) scores from the Meta Audiobox Aesthet- ics model [36] on both scenarios. The proposed model achieves the highest score in both and the highest overall mean. On STB, it shows the largest margin over all base- lines, while DMC ranks second and MEGAMI scores be- l...

  7. [7]

    CONCLUSION We proposed sequential stem blending as a principled reformulation of automatic music mixing, demonstrating that existing parallelized approaches are inherently not de- signed for this task while our model achieves strong stem blending performance and competitive results on the full AMM task. Beyond performance, the sequential paradigm naturall...

  8. [8]

    ACKNOWLEDGEMENTS The work is supported by grants from Google Asia Pacific, the National Science and Technology Council of Taiwan (NSTC 114-2628-E-002-013-MY3), and the Ministry of Education (MOE) of Taiwan (for Taiwan Centers of Ex- cellence in Artificial Intelligence)

Show all 44 references
  1. [9]

    De Man, R

    B. De Man, R. Stables, and J. D. Reiss,Intelligent mu- sic production. Focal Press, 2019

  2. [10]

    A deep learning approach to intelligent drum mixing with the wave-u-net,

    M. Martinez Ramirez, D. Stoller, and D. Moffat, “A deep learning approach to intelligent drum mixing with the wave-u-net,”Journal of the Audio Engineering So- ciety, vol. 69, 2021

  3. [11]

    Automatic music mixing with deep learning and out-of-domain data,

    M. A. Martínez-Ramírez, W.-H. Liao, G. Fabbro, S. Uhlich, C. Nagashima, and Y . Mitsufuji, “Automatic music mixing with deep learning and out-of-domain data,” inProc. International Society for Music Infor- mation Retrieval (ISMIR), 2022

  4. [12]

    Au- tomatic multitrack mixing with a differentiable mixing console of neural audio effects,

    C. J. Steinmetz, J. Pons, S. Pascual, and J. Serrà, “Au- tomatic multitrack mixing with a differentiable mixing console of neural audio effects,” inProc. International Conference on Acoustics, Speech, and Signal Process- ing (ICASSP), 2021

  5. [13]

    Automatic music mixing using a generative model of effect embeddings,

    E. Moliner, M. A. Martínez-Ramírez, J. Koo, W.-H. Liao, K. W. Cheuk, J. Serrà, V . Välimäki, and Y . Mit- sufuji, “Automatic music mixing using a generative model of effect embeddings,” inProc. International Conference on Acoustics, Speech, and Signal Process- ing (ICASSP), 2026

  6. [14]

    Ddsp: Differentiable digital signal processing,

    J. Engel, L. Hantrakul, C. Gu, and A. Roberts, “Ddsp: Differentiable digital signal processing,” inProc. In- ternational Conference on Learning Representations (ICLR), 2020

  7. [15]

    Diffvox: A differentiable model for capturing and analysing vocal effects distributions,

    C.-Y . Yu, M. A. Martínez-Ramírez, J. Koo, B. Hayes, W.-H. Liao, G. Fazekas, and Y . Mitsufuji, “Diffvox: A differentiable model for capturing and analysing vocal effects distributions,” inInternational Conference on Digital Audio Effects (DAFx), 2025

  8. [16]

    Search- ing for music mixing graphs: A pruning approach,

    S. Lee, M. A. Martínez-Ramírez, W.-H. Liao, S. Uh- lich, G. Fabbro, K. Lee, and Y . Mitsufuji, “Search- ing for music mixing graphs: A pruning approach,” in 27th International Conference on Digital Audio Effects (DAFx), 2024

  9. [17]

    Ddsp guitar amp: In- terpretable guitar amplifier modeling,

    Y .-T. Yeh, Y .-H. Chen, Y .-C. Cheng, J.-T. Wu, J.-J. Fu, Y .-F. Yeh, and Y .-H. Yang, “Ddsp guitar amp: In- terpretable guitar amplifier modeling,” inProc. Inter- national Conference on Acoustics, Speech, and Signal Processing (ICASSP), 2025

  10. [18]

    Deep learning for black-box modeling of audio ef- fects,

    M. A. Martínez Ramírez, E. Benetos, and J. D. Reiss, “Deep learning for black-box modeling of audio ef- fects,”Applied Sciences, vol. 10, 2020

  11. [19]

    Efficient neural net- works for real-time modeling of analog dynamic range compression,

    C. J. Steinmetz and J. D. Reiss, “Efficient neural net- works for real-time modeling of analog dynamic range compression,”152nd Convention of the Audio Engi- neering Society, 2021

  12. [20]

    Real time emula- tion of parametric guitar tube amplifier with long short term memory neural network,

    T. Schmitz and J.-J. Embrechts, “Real time emula- tion of parametric guitar tube amplifier with long short term memory neural network,”arXiv preprint arXiv:1804.07145, 2018

  13. [21]

    Hyper recur- rent neural network: Condition mechanisms for black- box audio effect modeling,

    Y .-T. Yeh, W.-Y . Hsiao, and Y .-H. Yang, “Hyper recur- rent neural network: Condition mechanisms for black- box audio effect modeling,” inInternational Confer- ence on Digital Audio Effects (DAFx), 2024

  14. [22]

    Flow matching for generative modeling,

    Y . Lipman, R. T. Chen, H. Ben-Hamu, M. Nickel, and M. Le, “Flow matching for generative modeling,” arXiv preprint arXiv:2210.02747, 2022

  15. [23]

    Flow straight and fast: Learning to generate and transfer data with rectified flow,

    X. Liu, C. Gong, and Q. Liu, “Flow straight and fast: Learning to generate and transfer data with rectified flow,” inProc. International Conference on Learning Representations (ICLR), 2023

  16. [24]

    Musicflow: Cascaded flow matching for text guided music generation,

    K. Prajwal, B. Shi, M. Lee, A. Vyas, A. Tjandra, M. Luthra, B. Guo, H. Wang, T. Afouras, D. Kantet al., “Musicflow: Cascaded flow matching for text guided music generation,” inProc. ICML, 2024

  17. [25]

    Stemphonic: All-at-once flexible multi- stem music generation,

    S.-L. Wu, G. Zhu, J.-P. Caceres, C.-Z. A. Huang, and N. J. Bryan, “Stemphonic: All-at-once flexible multi- stem music generation,” inProc. International Con- ference on Acoustics, Speech, and Signal Processing (ICASSP), 2026

  18. [26]

    Sam audio: Segment anything in audio,

    B. Shi, A. Tjandra, J. Hoffman, H. Wang, Y .-C. Wu, L. Gao, J. Richter, M. Le, A. Vyas, S. Chenet al., “Sam audio: Segment anything in audio,”arXiv preprint arXiv:2512.18099, 2025

  19. [27]

    Flowsep: Language-queried sound separation with rectified flow matching,

    Y . Yuan, X. Liu, H. Liu, M. D. Plumbley, and W. Wang, “Flowsep: Language-queried sound separation with rectified flow matching,” inProc. International Con- ference on Acoustics, Speech, and Signal Processing (ICASSP), 2025

  20. [28]

    Flowing from words to pixels: A noise-free frame- work for cross-modality evolution,

    Q. Liu, X. Yin, A. Yuille, A. Brown, and M. Singh, “Flowing from words to pixels: A noise-free frame- work for cross-modality evolution,” inProceedings of the Computer Vision and Pattern Recognition Confer- ence, 2025

  21. [29]

    Sonicmaster: Towards controllable all-in- one music restoration and mastering,

    J. Melechovsky, A. Mehrish, A. Roy, and D. Her- remans, “Sonicmaster: Towards controllable all-in- one music restoration and mastering,”arXiv preprint arXiv:2508.03448, 2025

  22. [30]

    Stable audio open,

    Z. Evans, J. D. Parker, C. Carr, Z. Zukowski, J. Taylor, and J. Pons, “Stable audio open,” inProc. International Conference on Acoustics, Speech, and Signal Process- ing (ICASSP), 2025

  23. [31]

    MedleyDB: A multitrack dataset for annotation-intensive mir research

    R. M. Bittner, J. Salamon, M. Tierney, M. Mauch, C. Cannam, and J. P. Bello, “MedleyDB: A multitrack dataset for annotation-intensive mir research.” inProc. International Society for Music Information Retrieval (ISMIR), 2014

  24. [32]

    Moisesdb: A dataset for source separation beyond 4- stems,

    I. Pereira, F. Araújo, F. Korzeniowski, and R. V ogl, “Moisesdb: A dataset for source separation beyond 4- stems,”arXiv preprint arXiv:2307.15913, 2023

  25. [33]

    Pyrooma- coustics: A python package for audio room simula- tion and array processing algorithms,

    R. Scheibler, E. Bezzam, and I. Dokmani ´c, “Pyrooma- coustics: A python package for audio room simula- tion and array processing algorithms,” inProc. Inter- national Conference on Acoustics, Speech, and Signal Processing (ICASSP), 2018

  26. [34]

    Scaling rectified flow transformers for high-resolution image synthesis,

    P. Esser, S. Kulal, A. Blattmann, R. Entezari, J. Müller, H. Saini, Y . Levi, D. Lorenz, A. Sauer, F. Boeselet al., “Scaling rectified flow transformers for high-resolution image synthesis,” inProc. ICML, 2024

  27. [35]

    Scalable diffusion models with transformers,

    W. Peebles and S. Xie, “Scalable diffusion models with transformers,” inProceedings of the IEEE/CVF inter- national conference on computer vision, 2023

  28. [36]

    Classifier-free diffusion guid- ance,

    J. Ho and T. Salimans, “Classifier-free diffusion guid- ance,”arXiv preprint arXiv:2207.12598, 2022

  29. [37]

    dasp-pytorch,

    C. J. Steinmetz, “dasp-pytorch,” [Online] https:// github.com/csteinmetz1/dasp-pytorch/

  30. [38]

    Diff-MST: Differentiable mixing style transfer,

    S. S. Vanka, C. Steinmetz, J.-B. Rolland, J. Reiss, and G. Fazekas, “Diff-MST: Differentiable mixing style transfer,” inProc. International Society for Music In- formation Retrieval (ISMIR), 2024

  31. [39]

    Kad: No more fad! an effective and efficient evaluation metric for audio generation,

    Y . Chung, P. Eu, J. Lee, K. Choi, J. Nam, and B. S. Chon, “Kad: No more fad! an effective and efficient evaluation metric for audio generation,” inProc. ICML, 2025

  32. [40]

    Fr\’echet audio distance: A metric for evalu- ating music enhancement algorithms,

    K. Kilgour, M. Zuluaga, D. Roblek, and M. Shar- ifi, “Fr\’echet audio distance: A metric for evalu- ating music enhancement algorithms,”arXiv preprint arXiv:1812.08466, 2018

  33. [41]

    Large-scale contrastive language- audio pretraining with feature fusion and keyword- to-caption augmentation,

    Y . Wu, K. Chen, T. Zhang, Y . Hui, T. Berg-Kirkpatrick, and S. Dubnov, “Large-scale contrastive language- audio pretraining with feature fusion and keyword- to-caption augmentation,” inProc. International Con- ference on Acoustics, Speech, and Signal Processing (ICASSP), 2023

  34. [42]

    Fx-encoder++: Extracting instrument-wise audio effects representa- tions from mixtures,

    Y .-T. Yeh, J. Koo, M. A. Martínez-Ramírez, W.-H. Liao, Y .-H. Yang, and Y . Mitsufuji, “Fx-encoder++: Extracting instrument-wise audio effects representa- tions from mixtures,” inProc. International Society for Music Information Retrieval (ISMIR), 2025

  35. [43]

    webmushra—a comprehensive framework for web-based listening tests,

    M. Schoeffler, S. Bartoschek, F.-R. Stöter, M. Roess, S. Westphal, B. Edler, and J. Herre, “webmushra—a comprehensive framework for web-based listening tests,”Journal of open research software, 2018

  36. [44]

    Meta audiobox aesthetics: Unified automatic qual- ity assessment for speech, music, and sound,

    A. Tjandra, Y .-C. Wu, B. Guo, J. Hoffman, B. Ellis, A. Vyas, B. Shi, S. Chen, M. Le, N. Zacharovet al., “Meta audiobox aesthetics: Unified automatic qual- ity assessment for speech, music, and sound,”arXiv preprint arXiv:2502.05139, 2025

Pith tools

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