Pith. sign in

REVIEW 2 major objections 6 minor 45 references

Teaching Speech Enhancement Models to Sing: Domain Adaptation from Speech Enhancement to Singing Voice Separation

T0 review · 2 major / 6 minor · reviewed 2026-07-14 · grok-4.5

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

Pith's one-line read Adapting speech-enhancement models to singing voice separation beats training from scratch when music data is scarce.

desk verdict Clean empirical recipe showing SE pretraining helps SVS under data scarcity; LoRA preserves the source task, but checkpoint selection on the reported test set softens the exact dB claims. read the letter →

arxiv 2607.11630 v1 pith:UGNQKCH7 submitted 2026-07-13 cs.SD eess.AS

Paul A. Bereuter , Mark D. Plumbley , Alois Sontacchi This is my paper
classification cs.SDeess.AS
keywords singingvoiceseparationspeechenhancementdomainadaptationlow-rankLoRAscore-basedgenerativemodelsband-splitRNNdatascarcity
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

Singing voice separation has far less labeled training data than speech enhancement. This paper treats the former as a domain-adaptation problem from the latter: start from models already trained on large speech-enhancement corpora, then fine-tune them on the limited music stems that exist. Both full fine-tuning and low-rank adapters improve separation quality over identical architectures trained from scratch, with gains of roughly 0.3–1.8 dB in signal-to-distortion ratio. Full fine-tuning reaches the highest singing scores but erases the original speech-enhancement skill; LoRA keeps that skill intact while still delivering competitive singing performance and adding only a few percent extra parameters. The generative model also generalizes better to an unseen restoration test set. The practical claim is that speech-enhancement pretraining is a useful and data-efficient route into singing separation.

What carries the argument

Domain adaptation of two pretrained SE architectures (discriminative BSRNN and generative score-based SGM) via either full-parameter fine-tuning or LoRA low-rank adapters on MUSDB18-HQ + MoisesDB, evaluated for the trade-off between target-domain SVS/SVR performance and source-domain SE retention.

What would settle it

Train identical architectures from scratch on the same music data and show that the adapted models no longer improve SDR (or the other reported metrics) on the GenSVS and MSRBench test sets relative to those from-scratch baselines.

Watch

Extended reading notes

Core claim

Adapting pretrained speech-enhancement models—either by full fine-tuning or by LoRA—to singing voice separation consistently outperforms the same architectures trained from scratch on the available music data (0.29–1.8 dB SDR), establishing SE pretraining as an effective strategy under data scarcity. Full fine-tuning maximizes singing scores at the cost of catastrophic forgetting of speech enhancement; LoRA recovers most of the singing gain while preserving the original speech capability with only 6–12 % extra parameters.

Load-bearing premise

That the features learned on large speech-enhancement corpora transfer well enough to music-production processing and accompaniment that a few dozen hours of music stems can successfully adapt them.

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

2 major / 6 minor

Summary. The paper reformulates singing voice separation (SVS) as domain adaptation from speech enhancement (SE). It adapts a discriminative BSRNN (pretrained on ~700 h URGENT) and a generative score-based model (SGMSE, pretrained on ~87 h EARS-WHAM) to the limited ~35 h MUSDB18-HQ + MoisesDB data via full fine-tuning and LoRA. Both strategies outperform identical architectures trained from scratch by 0.29–1.8 dB SDR on GenSVS; full fine-tuning is strongest on SVS/SVR but degrades SE (PESQ drops), while LoRA (6–12 % extra parameters at modest ranks) retains SE performance and remains competitive. The generative model shows larger GenSVS-to-MSRBench gains, supporting better out-of-domain generalization for singing voice restoration. The authors conclude that SE pretraining is an effective strategy under data scarcity.

Significance. If the transfer claim holds under proper model selection, the work supplies a practical, low-overhead route to competitive SVS models without large music corpora, and shows that LoRA can preserve multi-domain capability—directly useful for universal source-separation systems. Strengths include a clean experimental design (same architectures from-scratch vs. adapted, multiple LoRA ranks, discriminative and generative families, three evaluation regimes with standard and embedding-based metrics), open code and audio examples, and an explicit trade-off analysis (Table 1, Figure 1, Table 3). The contribution is empirical rather than theoretical, but the protocol and public artifacts make the result reproducible and immediately actionable for the community.

major comments (2)
  1. §3.4 states that “Checkpoints with the highest Signal-to-Distortion-Ratio (SDR) on the GenSVS test set are selected for testing.” GenSVS is the sole in-domain SVS evaluation set that supplies the headline 0.29–1.8 dB gains (abstract, Table 1). Selecting the best epoch on the reported test set itself optimistically biases every adapted and from-scratch entry, so the quantitative support for the transfer claim may shrink or reverse once a held-out validation split is used for model selection. A re-run with proper validation-based checkpointing (or at least reporting of the validation protocol) is required before the central claim can be accepted at face value.
  2. The two architectures are pretrained on substantially different corpora (URGENT ~700 h vs. EARS-WHAM ~87 h) and different STFT configurations (N_fft = 960 vs. 1534). Table 2 and the generalization discussion in §4 attribute superior out-of-domain behavior primarily to the generative modeling paradigm. Because pretraining data volume and spectral front-end are confounded with architecture class, the claim that “the generative model shows improved generalization” cannot be cleanly isolated. Either a controlled ablation (same pretraining data / same STFT) or a clearer caveat is needed.
minor comments (6)
  1. Index terms and §2.2: “low rank adaption” / “Low-Rank adaptation” should be consistently “adaptation”.
  2. §3.2: “5 sexcerpts” is a typographical error for “5 s excerpts”.
  3. Table 1 caption and body: clarify that “LoRA 16/32/128” denote rank r; the current notation is slightly ambiguous for readers skimming the table.
  4. Figure 1: the two panels share a common PESQ axis but different MERT-MSE ranges; a brief note that lower-right is better would improve readability.
  5. §3.4: the statement that r = 8 “was tested but its separation performance was insufficient” should be quantified (e.g., SDR) so that the rank-selection decision is reproducible.
  6. References [13] and [21] are the authors’ own prior metric / Mel-RoFormer work; a short sentence distinguishing the present contribution from those baselines would help readers unfamiliar with the lineage.

Circularity Check

1 steps flagged · score 1.0 of 10

Empirical transfer study with no definitional loop; only minor evaluation leakage from selecting checkpoints on the reported GenSVS test set itself.

  1. fitted input called prediction [§3.4 (Model configurations / training protocol)]
    "Checkpoints with the highest Signal-to-Distortion-Ratio (SDR) on the GenSVS test set are selected for testing."

    The headline SDR numbers (and therefore the 0.29–1.8 dB deltas) on the sole in-domain SVS set are obtained by choosing the epoch that maximises that exact metric on the same set. Absolute performance is therefore the training-trajectory maximum rather than an independent evaluation; the reported figures are fitted to the test data by construction. Relative ranking among models may still be informative, but the quantitative claim is optimistically biased.

full rationale

The paper is a purely empirical domain-adaptation study. Its central claim (adapted SE models beat from-scratch counterparts by 0.29–1.8 dB SDR under data scarcity) rests on measured performance of trained networks against held-out stems, not on any equation that reduces a predicted quantity to a fitted free parameter or to a self-cited uniqueness theorem. The two architectures, the LoRA construction, the loss functions, and the three evaluation sets are all externally specified; the only soft spot is the protocol sentence that selects the checkpoint maximizing SDR on GenSVS and then reports that same SDR. This is optimistic bias / evaluation leakage, not a circular derivation of the transfer hypothesis. Self-citations are confined to the authors’ prior metric papers and the GenSVS excerpt definition; they do not underwrite the success criterion. Consequently the circularity burden is negligible (score 1).

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

The central claim rests on standard supervised fine-tuning assumptions plus the empirical hypothesis that SE representations transfer to SVS. Free parameters are ordinary training hyper-parameters and LoRA ranks chosen by the authors; no new physical constants or invented mediators appear. Axioms are ordinary domain-adaptation and deep-learning background.

free parameters (4)
  • LoRA rank r = 16/32/128 (BSRNN), 16 (SGM)
    Treated as a hyper-parameter; values {16,32,128} for BSRNN and fixed r=16 for SGM are chosen by the authors and directly affect reported parameter counts and performance.
  • LoRA scaling factor alpha = 2r or 32
    Set heuristically to alpha=2r (BSRNN) or alpha=32 (SGM); controls adapter contribution strength.
  • learning rates and batch sizes = various (1e-4, 2e-4, 5e-5)
    Chosen per model family and adaptation style (e.g., 5e-5 / batch 14 for fine-tuned BSRNN); affect convergence and final SDR.
  • number of diffusion steps and corrector steps = 45 / 0.5 / 2
    Fixed to N=45 steps, step size 0.5, 2 ALD corrections for all SGM inference; inherited from prior work but still free choices that influence generative metrics.
assumptions (3)
  • domain assumption Representations learned on large-scale speech-enhancement data transfer usefully to singing-voice mixtures under music-production degradations.
    Core hypothesis stated in §1; without it the fine-tuning protocol would not be expected to outperform training from scratch.
  • domain assumption Standard supervised losses (weighted L1 + SI-SDR for BSRNN; conditional denoising score-matching for SGM) remain appropriate after domain shift.
    Losses are taken unchanged from the SE pretraining literature (§3.3).
  • standard math LoRA low-rank updates (Eq. 3) can capture the residual domain gap while leaving the frozen base weights intact.
    Standard LoRA construction from Hu et al. (2022), applied without modification.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Teaching Speech Enhancement Models to Sing: Domain Adaptation from Speech Enhancement to Singing Voice Separation." pith.science (2026). https://pith.science/paper/UGNQKCH7

@misc{pith2026260711630,
  author       = {Pith},
  title        = {Pith review of: Teaching Speech Enhancement Models to Sing: Domain Adaptation from Speech Enhancement to Singing Voice Separation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UGNQKCH7}},
  note         = {Machine review of arXiv:2607.11630}
}
read the original abstract

State-of-the-art speech enhancement models benefit from large-scale labeled datasets, whereas singing voice separation models suffer from limited available training data. To address this limitation, we formulate singing voice separation as domain adaptation from speech enhancement to singing voice separation. We investigate two fine-tuning strategies: full fine-tuning and parameter-efficient fine-tuning using Low-Rank Adaptation (LoRA) on a discriminative and a generative model. Models with either adaptation strategy outperform the same architectures trained from scratch by 0.29-1.8 dB in Signal-to-Distortion-Ratio. Full fine-tuning yields the highest singing voice separation performance, but catastrophic forgetting degrades speech enhancement performance. LoRA fine-tuning achieves competitive singing voice separation performance while preserving the original speech enhancement capability with only 6-12% additional parameters compared to the base speech enhancement model. Furthermore, the generative model shows improved generalization to an unseen test set. The results demonstrate that adapting pretrained speech enhancement models is an effective strategy for training singing voice separation models in data-scarce scenarios.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

45 extracted references · 1 linked inside Pith

  1. [1]

    Interactive Audiovisual Digital Twins of Performance Venues

    INTRODUCTION Classical speech enhancement (SE) has primarily addressed acous- tic degradations such as environmental noise and reverberation, as tackled in organized challenges such as the Deep Noise Suppression (DNS) Challenge [1–3]. The recent transition to more universal en- hancement scenarios, such as the Universality, Robustness, and Gen- eralizabil...

  2. [2]

    Tasks related to separation of vocal signals SE and SVS can both be formulated as vocal signal separation prob- lems

    METHOD 2.1. Tasks related to separation of vocal signals SE and SVS can both be formulated as vocal signal separation prob- lems. In universal SE settings, the observed discrete-time mixture signaly[n]can be written as: y[n] =D s s[n] +n[n](1) arXiv:2607.11630v1 [cs.SD] 13 Jul 2026 wheres[n]denotes the clean speech signal,D s(·)represents a po- tentially ...

    arXiv 2026

  3. [3]

    Full fine-tuning With full fine-tuning all parameters of the pretrained SE model are updated during SVS training

    combined. Full fine-tuning With full fine-tuning all parameters of the pretrained SE model are updated during SVS training. While this approach allows the model to re-learn features which better match the singing voice domain, a performance degradation in the SE domain due to catastrophic for- getting [16] can be expected [17]. Low-Rank adaptation Origina...

  4. [4]

    Pretraining and adaptation All models in this work operate at a sampling rate of48 kHzand are initialized from pretrained checkpoints

    EXPERIMENTAL SETUP 3.1. Pretraining and adaptation All models in this work operate at a sampling rate of48 kHzand are initialized from pretrained checkpoints. For the BSRNN, weights pretrained on the URGENT dataset (approx.700 hof audio) from

  5. [5]

    The SGM, which utilizes a noise-conditional score network (NCSN++) [22] as its backbone, was pretrained for SE on the EARS-WHAM dataset [23] (approx.87 h)

    are employed. The SGM, which utilizes a noise-conditional score network (NCSN++) [22] as its backbone, was pretrained for SE on the EARS-WHAM dataset [23] (approx.87 h). 3.2. Datasets To adapt the models for the singing voice domain, we utilize two la- beled multi-stem datasets: MUSDB18-HQ [9] and MoisesDB [10]. These datasets provide isolated vocal and a...

  6. [6]

    from scratch

    RESULTS AND DISCUSSION Performance trade-off in domain adaptation As shown in Table 1 and Figure 1, full fine-tuning achieves the highest scores across both singing voice datasets. On GenSVS (in- 0.100 0.110 0.120 0.130 0.140 0.150 0.160 0.170MERT-MSE full scratch base LoRA 16 base LoRA 16 LoRA 32/128 scratch full GenSVS/EARS-WHAM SGM BSRNN 1.75 2.00 2.25...

  7. [7]

    While full fine-tuning achieves the best SVS performance, our results show that LoRA enables domain adaptation while preserving the original model capabilities

    CONCLUSION In this work, we investigate the adaptation of pretrained speech en- hancement (SE) models to singing voice separation (SVS). While full fine-tuning achieves the best SVS performance, our results show that LoRA enables domain adaptation while preserving the original model capabilities. Leveraging its additive structure, LoRA retains the origina...

  8. [8]

    INTERSPEECH 2021 deep noise suppression challenge,

    C. K. Reddy, H. Dubey, K. Koishida, A. Nair, V . Gopal, R. Cutler, S. Braun, H. Gamper, R. Aichner, and S. Srinivasan, “INTERSPEECH 2021 deep noise suppression challenge,” inProc. Interspeech, 2021, pp. 2796–2800

    2021

Show all 45 references
  1. [9]

    ICASSP 2022 deep noise suppression challenge,

    H. Dubey, V . Gopal, R. Cutler, A. Aazami, S. Matusevych, S. Braun, S. E. Eskimez, M. Thakker, T. Yoshioka, H. Gamper, and R. Aichner, “ICASSP 2022 deep noise suppression challenge,” inProc. ICASSP, 2022, pp. 9271–9275

    2022

  2. [10]

    ICASSP 2023 deep noise suppression challenge,

    H. Dubey, A. Aazami, V . Gopal, B. Naderi, S. Braun, R. Cutler, A. Ju, M. Zohourian, M. Tang, M. Golestaneh, and R. Aichner, “ICASSP 2023 deep noise suppression challenge,”IEEE Open J. of Signal Pro- cess., vol. 5, pp. 725–737, 2024

    2023

  3. [11]

    URGENT challenge: Universality, robustness, and generalizability for speech en- hancement,

    W. Zhang, R. Scheibler, K. Saijo, S. Cornell, C. Li, Z. Ni, J. Pirkl- bauer, M. Sach, S. Watanabe, T. Fingscheidt, and Y . Qian, “URGENT challenge: Universality, robustness, and generalizability for speech en- hancement,” inProc. Interspeech, 2024, pp. 4868–4872

    2024

  4. [12]

    Speech enhancement and dereverberation with diffusion-based gen- erative models,

    J. Richter, S. Welker, J.-M. Lemercier, B. Lay, and T. Gerkmann, “Speech enhancement and dereverberation with diffusion-based gen- erative models,”IEEE/ACM Trans. Audio, Speech, Lang. Process., vol. 31, pp. 2351–2364, 2023

    2023

  5. [13]

    StoRM: A diffusion-based stochastic regeneration model for speech enhance- ment and dereverberation,

    J.-M. Lemercier, J. Richter, S. Welker, and T. Gerkmann, “StoRM: A diffusion-based stochastic regeneration model for speech enhance- ment and dereverberation,”IEEE/ACM Trans. Audio, Speech, Lang. Process., vol. 31, pp. 2724–2737, 2023

    2023

  6. [14]

    Universal score- based speech enhancement with high content preservation,

    R. Scheibler, Y . Fujita, Y . Shirahata, and T. Komatsu, “Universal score- based speech enhancement with high content preservation,” inProc. Interspeech, 2024, pp. 1165–1169

    2024

  7. [15]

    Diffusion-based signal refiner for speech enhancement and separa- tion,

    M. Hirano, R. Sawata, N. Murata, S. Takahashi, and Y . Mitsufuji, “Diffusion-based signal refiner for speech enhancement and separa- tion,”IEEE Trans. Audio, Speech, Lang. Process., vol. 34, pp. 1426– 1440, 2026

    2026

  8. [16]

    The MUSDB18 corpus for music separation,

    Z. Rafii, A. Liutkus, F.-R. St ¨oter, S. I. Mimilakis, and R. Bittner, “The MUSDB18 corpus for music separation,” Dec. 2017. [Online]. Available: https://doi.org/10.5281/zenodo.1117372

    2017 doi

  9. [17]

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

    I. Pereira, F. Ara ´ujo, F. Korzeniowski, and R. V ogl, “MoisesDB: A dataset for source separation beyond 4-stems,” inProc. ISMIR, 2023, pp. 619–626

    2023

  10. [18]

    High fidelity speech enhancement with band-split RNN,

    J. Yu, H. Chen, Y . Luo, R. Gu, and C. Weng, “High fidelity speech enhancement with band-split RNN,” inProc. Interspeech, 2023, pp. 2483–2487

    2023

  11. [19]

    Music source separation with band-split RNN,

    Y . Luo and J. Yu, “Music source separation with band-split RNN,” IEEE Trans. Audio, Speech, Lang. Process., vol. 31, pp. 1893–1901, 2023

    1901

  12. [20]

    Towards reliable objective evaluation metrics for generative singing voice sepa- ration models,

    P. A. Bereuter, B. Stahl, M. D. Plumbley, and A. Sontacchi, “Towards reliable objective evaluation metrics for generative singing voice sepa- ration models,” inProc. WASPAA, 2025

    2025

  13. [21]

    Diff-VS: Efficient audio-aware diffusion u-net for vocals separation,

    Y .-N. A. Hung, R. V ogl, F. Korzeniowski, and I. Pereira, “Diff-VS: Efficient audio-aware diffusion u-net for vocals separation,” inProc. ICASSP, 2026, pp. 14 867–14 871

    2026

  14. [22]

    Music source restora- tion,

    Y . Zang, Z. Dai, M. D. Plumbley, and Q. Kong, “Music source restora- tion,” inProc. MMSP, 2025, pp. 138–143

    2025

  15. [23]

    Overcoming catastrophic forgetting in neural networks,

    J. Kirkpatrick, R. Pascanu, N. Rabinowitz, J. Veness, G. Des- jardins, A. A. Rusu, K. Milan, J. Quan, T. Ramalho, A. Grabska-Barwinska, D. Hassabis, C. Clopath, D. Kumaran, and R. Hadsell, “Overcoming catastrophic forgetting in neural networks,”Proceedings of the National Acad...

    2017 doi

  16. [24]

    SERIL: Noise adaptive speech enhancement using regularization-based incre- mental learning,

    C.-C. Lee, Y .-C. Lin, H.-T. Lin, H.-M. Wang, and Y . Tsao, “SERIL: Noise adaptive speech enhancement using regularization-based incre- mental learning,” inProc. Interspeech, 2020, pp. 2432–2436

    2020

  17. [25]

    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,” inProc. ICLR, 2022. [Online]. Available: https://openreview.net/forum?id=nZeVKeeFYf9

    2022

  18. [26]

    Parameter- efficient transfer learning of audio spectrogram transformers,

    U. Cappellazzo, D. Falavigna, A. Brutti, and M. Ravanelli, “Parameter- efficient transfer learning of audio spectrogram transformers,” inProc. MLSP, 2024

    2024

  19. [27]

    Improving anomalous sound detection via low-rank adaptation fine- tuning of pre-trained audio models,

    X. Zheng, A. Jiang, B. Han, Y . Qian, P. Fan, J. Liu, and W.-Q. Zhang, “Improving anomalous sound detection via low-rank adaptation fine- tuning of pre-trained audio models,” inProc. SLT, 2024, pp. 969–974

    2024

  20. [28]

    Mel-band-roformer-vocal-model,

    K. Jensen, “Mel-band-roformer-vocal-model,” 2024, GitHub reposi- tory. [Online]. Available: https://github.com/KimberleyJensen/Mel- Band-Roformer-V ocal-Model

    2024

  21. [29]

    Score-based generative modeling through stochastic differential equations,

    Y . Song, J. Sohl-Dickstein, D. P. Kingma, A. Kumar, S. Ermon, and B. Poole, “Score-based generative modeling through stochastic differential equations,” inProc. ICLR, 2021. [Online]. Available: https://openreview.net/forum?id=PxTIG12RRHS

    2021

  22. [30]

    EARS: An anechoic fullband speech dataset benchmarked for speech enhancement and dereverberation,

    J. Richter, Y .-C. Wu, S. Krenn, S. Welker, B. Lay, S. Watanabe, A. Richard, and T. Gerkmann, “EARS: An anechoic fullband speech dataset benchmarked for speech enhancement and dereverberation,” in Proc. Interspeech, 2024, pp. 4873–4877

    2024

  23. [31]

    MSRBench: A benchmarking dataset for music source restoration,

    Y . Zang, J. Hai, W. Ge, Q. Kong, Z. Dai, H. Wang, Y . Mitsufuji, and M. D. Plumbley, “MSRBench: A benchmarking dataset for music source restoration,” 2025. [Online]. Available: https: //arxiv.org/abs/2510.10995

    2025

  24. [32]

    EBU R 128: Loudness normalisation and permitted maximum level of audio signals,

    European Broadcasting Union, “EBU R 128: Loudness normalisation and permitted maximum level of audio signals,” Geneva, Switzerland, Nov. 2023

    2023

  25. [33]

    Generative modeling by estimating gradients of the data distribution,

    Y . Song and S. Ermon, “Generative modeling by estimating gradients of the data distribution,” inProc. NeurIPS, vol. 32. Curran Associates, Inc., 2019. [Online]. Available: https://proceedings.neurips.cc/paper files/paper/2019/file/3001ef257407d5a371a96dcd947c7d93 - Paper.pdf

    2019

  26. [34]

    PEFT: State-of-the-art parameter-efficient fine- tuning methods,

    S. Mangrulkar, S. Gugger, L. Debut, Y . Belkada, S. Paul, B. Bossan, and M. Tietz, “PEFT: State-of-the-art parameter-efficient fine- tuning methods,” https://github.com/huggingface/peft, 2026, version 0.18.1

    2026

  27. [35]

    Delving deep into rectifiers: Sur- passing human-level performance on imagenet classification,

    K. He, X. Zhang, S. Ren, and J. Sun, “Delving deep into rectifiers: Sur- passing human-level performance on imagenet classification,” inProc. ICCV, 2015, pp. 1026–1034

    2015

  28. [36]

    Performance measurement in blind audio source separation,

    E. Vincent, R. Gribonval, and C. Fevotte, “Performance measurement in blind audio source separation,”IEEE Trans. Audio, Speech, Lang. Process., vol. 14, no. 4, pp. 1462–1469, 2006

    2006

  29. [37]

    TorchMetrics - measuring reproducibility in PyTorch,

    N. S. Detlefsen, J. Borovec, J. Schock, A. Harsh, T. Koker, L. D. Liello, D. Stancl, C. Quan, M. Grechkin, and W. Falcon, “TorchMetrics - measuring reproducibility in PyTorch,” 2 2025T, version 1.8.2. [Online]. Available: https://www.pytorchlightning.ai

  30. [38]

    MERT: Acous- tic music understanding model with large-scale self-supervised train- ing,

    Y . LI, R. Yuan, G. Zhang, Y . Ma, X. Chen, H. Yin, C. Xiao, C. Lin, A. Ragni, E. Benetos, N. Gyenge, R. Dannenberg, R. Liu, W. Chen, G. Xia, Y . Shi, W. Huang, Z. Wang, Y . Guo, and J. Fu, “MERT: Acous- tic music understanding model with large-scale self-supervised train- ing...

    2024

  31. [39]

    auraloss: Audio focused loss functions in PyTorch,

    C. J. Steinmetz and J. D. Reiss, “auraloss: Audio focused loss functions in PyTorch,” inDigital music research network one-day workshop (DMRN+15), 2020, version 0.4.0. [Online]. Available: https://pypi.org/project/auraloss/

    2020

  32. [40]

    Embedding-based intrusive evaluation metrics for musical source separation using MERT representations,

    P. A. Bereuter, “Embedding-based intrusive evaluation metrics for musical source separation using MERT representations,” inF ortschritte der Akustik – DAGA 2026. Berlin, Germany: German Acoustical Society (DEGA e.V .), 2026, pp. 943–946. [Online]. Available: https: //pub.dega-...

    2026

  33. [41]

    SDR – Half- baked or well done?

    J. L. Roux, S. Wisdom, H. Erdogan, and J. R. Hershey, “SDR – Half- baked or well done?” inProc. ICASSP, 2019, pp. 626–630

    2019

  34. [42]

    Perceptual evaluation of speech quality (pesq)-a new method for speech quality assessment of telephone networks and codecs,

    A. Rix, J. Beerends, M. Hollier, and A. Hekstra, “Perceptual evaluation of speech quality (pesq)-a new method for speech quality assessment of telephone networks and codecs,” inProc. ICASSP, vol. 2, 2001, pp. 749–752 vol.2

    2001

  35. [43]

    Distillation and pruning for scalable self- supervised representation-based speech quality assessment,

    B. Stahl and H. Gamper, “Distillation and pruning for scalable self- supervised representation-based speech quality assessment,” inProc. ICASSP, 2025

    2025

  36. [44]

    Mel-RoFormer for vocal separa- tion and vocal melody transcription,

    J.-C. Wang, W.-T. Lu, and J. Chen, “Mel-RoFormer for vocal separa- tion and vocal melody transcription,” inProc. ISMIR. ISMIR, Nov. 2024, pp. 454–461

    2024

  37. [45]

    ptflops: A flops counting tool for neural networks in pytorch framework,

    V . Sovrasov, “ptflops: A flops counting tool for neural networks in pytorch framework,” 2025, version 0.7.5. [Online]. Available: https://github.com/sovrasov/flops-counter.pytorch

    2025

Pith tools

Reviewed July 14, 2026 · model on record in the stance chip above.