Pith. sign in

REVIEW 4 major objections 9 minor 27 references

MambaRate: Speech Quality Assessment Across Different Sampling Rates

T0 review · 4 major / 9 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 MambaRate shows that a 900k-parameter model using frozen WavLM embeddings and Mamba-2 layers can match or beat a fine-tuned wav2vec MOS baseline in a few-shot setting, ranking fourth of five in AudioMOS 2025 Track 3.

desk verdict A competent challenge paper with a real held-out result, but the sampling-rate robustness headline is undercut by its own 16 kHz front-end. read the letter →

arxiv 2507.12090 v1 pith:Z7CBWIZ5 submitted 2025-07-16 cs.SD eess.AS

Panos Kakoulidis , Iakovi Alexiou , Junkwang Oh , Gunu Jho , Inchul Hwang , Pirros Tsiakoulis , Aimilios Chalamandaris This is my paper
classification cs.SDeess.AS
keywords MOSpredictionspeechqualityassessmentsamplingraterobustnessstatespacemodelsWavLMembeddingsGaussianradialbasisfunctionsfew-shotlearningAudiochallenge
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

The paper tries to establish that accurate MOS prediction across different sampling rates does not require a large fine-tuned SSL model. MambaRate uses frozen WavLM embeddings, a small convolutional downsampler, and Mamba-2 state space layers, totaling about 900k trainable parameters. In the AudioMOS 2025 Track 3 few-shot setting, its T16 version beat the fine-tuned wav2vec baseline (B03) by roughly 14% in SRCC without any pretraining on MOS data, and finished fourth of five, about 6% behind the winner. The paper also reports ablations showing that embeddings from other WavLM layers or from Whisper improve on T16. If this holds, lightweight MOS predictors can be trained in minutes on small data while generalizing across 16, 24, and 48 kHz speech.

What carries the argument

The architecture combines frozen WavLM-Large embeddings (layer 6 for T16) with a 1D convolutional downsampling block, several Mamba-2 blocks (a selective state space model that processes sequences efficiently), and a two-layer MLP head with sigmoid output. The central encoding trick is that each MOS rating is converted into a 16-dimensional vector via Gaussian radial basis functions centered between 1 and 5, and the model is trained with MSE to predict that vector; at inference, the scalar prediction is the center with the maximum value. This avoids treating MOS as a plain regression target and lets the model represent rating uncertainty continuously. The whole model is trained from scratch on the challenge's few hundred clips with AdamW, cosine annealing, and early stopping in well under 30 epochs.

What would settle it

Take a set of 24 kHz or 48 kHz utterances, create two versions of each (one full-band and one low-pass filtered at 8 kHz), and collect human MOS for both. If human scores separate the pairs while MambaRate, which sees only 16 kHz audio, gives identical predictions for both, then the model has not actually learned to be sampling-rate invariant; it has learned to ignore the missing band. A simpler check is to compute MambaRate's output before and after downsampling and see whether it changes at all.

Watch

Extended reading notes

Core claim

The paper's central claim is that a compact predictor about 900k parameters in size can predict MOS for speech recorded at 16, 24, and 48 kHz almost as well as or better than a much larger fine-tuned wav2vec 2.0 baseline, provided the input is normalized to 16 kHz and represented by frozen self-supervised embeddings. Specifically, MambaRate's challenged T16 system achieved a system-level SRCC about 14% higher than the baseline in the few-shot Track 3 setting, ranking fourth out of five systems and coming within roughly 6% of the winning system. On the BVCC 16 kHz benchmark, MambaRate matched the baseline in LCC and SRCC while improving MSE and Kendall's tau. Additional experiments with mean ratings and embeddings from later WavLM layers or from Whisper outperform the submitted T16, supporting the paper's conclusion that offline pre-trained representations can be efficiently adopted for few-shot MOS prediction across sampling rates.

Load-bearing premise

The claim depends on the assumption that downsampling every input to 16 kHz keeps all the audible information that drives human quality ratings, so any quality differences tied to frequencies above 8 kHz are either absent or not needed in the challenge data.

Editorial extensions

If this is right

  • A roughly 900k-parameter MOS predictor trained for minutes can beat a fine-tuned wav2vec baseline in a few-shot setting, reducing the compute needed for objective speech quality evaluation.
  • Frozen SSL embeddings from a single layer are sufficient input for competitive MOS prediction, and no listener or system metadata is required.
  • Pre-training on BVCC before fine-tuning on challenge data improves performance, raising system-level SRCC from 0.879 to 0.930.
  • Using embeddings from WavLM layer 9 or from Whisper outperforms the submitted layer-6 T16 version, suggesting that input representation choice matters more than added capacity.
  • The same design should transfer to other small-data MOS tasks, with future work planned on the SOMOS and VCC2018 datasets.

Reading between the lines

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

  • Because every input is downsampled to 16 kHz before scoring, MambaRate's sampling-rate robustness is really robustness to input normalization; any quality cue that lives only above 8 kHz is invisible to the model, so the claim should be tested on band-limited versus full-band versions of the same recordings.
  • The RBF target encoding may be the key reason the small model trains so quickly; a testable extension is to compare it against a plain scalar regression head with the same architecture.
  • Whisper embeddings, trained for ASR, beating WavLM embeddings suggests that quality-relevant information is largely captured by speech content representations, and ensembling multiple frozen layers could improve accuracy further.
  • A useful follow-up would measure at what training-set size the fine-tuned wav2vec baseline overtakes MambaRate, mapping where the few-shot advantage ends.
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

4 major / 9 minor

Summary. The paper presents MambaRate, a lightweight MOS prediction model that uses frozen WavLM-Large embeddings fed through Mamba-2 blocks, with target ratings encoded via Gaussian radial basis functions. The system was submitted to Track 3 of the AudioMOS 2025 Challenge, where the T16 version ranked fourth among five systems and is claimed to outperform the challenge baseline (B03) by about 14% in system-level SRCC. The paper also reports utterance-level comparisons against a UTMOS-style baseline on BVCC and challenge data, and ablation studies exploring different WavLM layers and alternative SSL embeddings (Whisper, VGGish). The manuscript emphasizes sampling-rate robustness as a central contribution.

Significance. If the empirical results hold, MambaRate demonstrates that a model with roughly 900k parameters, using offline SSL embeddings and no fine-tuning of the front-end, can compete with a fine-tuned wav2vec-2.0 baseline in a few-shot challenge setting, at a fraction of the training cost. The independent challenge leaderboard provides external validation for the T16 result, which is a clear strength. However, the significance of the paper's headline claim—"Speech Quality Assessment Across Different Sampling Rates"—is substantially undermined by the preprocessing choice described in Section III-A, where all audio is downsampled to 16 kHz before the model ever sees it. Also, the ablation comparisons are reported on the validation split used for model selection, and no statistical uncertainty is reported anywhere. As such, the paper's empirical value is real but modest, and its framing overstates what is actually demonstrated.

major comments (4)
  1. [III-A and Abstract/Conclusion] Section III-A states: "All audios are preprocessed ... and downsampled to 16kHz for compatibility with the pre-trained models in use." This means MambaRate never processes spectral content above 8 kHz, so the abstract's claim of "limited bias regarding the sampling rate" and the conclusion's "supporting different sampling rates" are not supported by the experiments. The model is sampling-rate-agnostic by input normalization, not by learned robustness against sampling-rate variation. Please reframe these claims to describe a 16 kHz front-end applied to multi-rate recordings, or provide evidence (e.g., per-sampling-rate metrics or an analysis of whether human quality ratings in Track 3 are insensitive to content above 8 kHz) that the downsampling does not hide the very differences the paper claims to address.
  2. [III-C, Table II] The evaluation protocol for Table II is underspecified: it is unclear whether the metrics are computed on the challenge test set, the 10% held-out split of the challenge training set, or the challenge development set. Since T16's leaderboard result is a system-level SRCC while Table II reports utterance-level metrics, the claimed ~14% improvement over B03 cannot be directly verified from the paper. Please state explicitly which partition is used for each column and provide the system-level SRCC values for T16 and B03 in the text (or in the figure caption).
  3. [III-D, Tables III and IV] The ablations in Tables III and IV are trained on a 70-15-15 split, and the reported metrics are computed on the same 15% validation split used for model selection (e.g., choosing the WavLM layer). This selection-on-validation procedure inflates the apparent differences and makes the statement that these ablations "outperform T16" unreliable. Please report results on a truly held-out partition, such as the 10% test split used for T16, or use repeated cross-validation with variance estimates.
  4. [All experiments] No error bars, confidence intervals, or significance tests are provided for any table. Differences between WavLM layers in Table III and between embedding types in Table IV are only a few hundredths in SRCC, which may well be within run-to-run variance for a model of this size. At minimum, report multiple seeded runs with standard deviations for the key comparisons (Tables II and III). Without this, the robustness of the reported improvements cannot be assessed.
minor comments (9)
  1. [Abstract] The abstract contains two typographical issues: "Correllation" should be "Correlation" and "fewshot" should be "few-shot."
  2. [II, Eq. (1)] The RBF width sigma in Eq. (1) is never defined; please report its value or describe how it is chosen, since it directly controls the shape of the target encoding.
  3. [II] The text says "an 1D convolutional layer" and "a multi-perceptron block"; these should be "a 1D convolutional layer" and "a multilayer perceptron (MLP) block" for clarity.
  4. [III-A] The SV56 amplitude normalization is mentioned without a reference; please cite the corresponding ITU-T recommendation or the implementation used.
  5. [Table II caption] The column headers ("BVCC", "BVCC+C", "C", "BVCC+C") make the comparison difficult to parse; clarify in the caption or table body which column corresponds to the baseline and which to MambaRate, and what each training condition means.
  6. [II (decoding)] The decoding from the 16 RBF outputs uses argmax over centers, which quantizes predictions to 16 discrete levels; a weighted average over centers would give a continuous prediction and might improve MSE without affecting the rank-based metrics.
  7. [III-C] The sentence "MambaRate clearly outperforms both setups of the baseline, although the former is trained only on the training data of the challenge" is ambiguous: identify whether "the former" refers to MambaRate or to one of the baseline setups.
  8. [II] The cosine annealing scheduler parameter "Tmax = 10" is not specified in units; clarify whether it is 10 epochs or 10 optimizer steps.
  9. [Figure 2] Figure 2 is reproduced without axis labels or numeric values; please provide the system-level SRCC values for all challenge systems in the caption or in the text.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the central performance claims are evaluated on externally held-out AudioMOS 2025 challenge data, and no load-bearing step reduces to its own inputs.

full rationale

The central claims are grounded in the AudioMOS 2025 Track 3 evaluation, which used unseen challenge test items with human MOS labels; the T16 model was trained only on the provided 400-item training set (Section III-A) and ranked fourth out of five on the official leaderboard (Figure 2). That is an external, non-circular benchmark. The BVCC comparisons use a separate public dataset and the baseline's published weights, so they are also independent. The only manuscript element that could look circular is the ablation study (Section III-D), where the WavLM layer and embedding model are compared on a 70-15-15 split and the best performer is then reported as a finding; however, this is standard model selection on a validation split, not a fitted parameter being repackaged as a prediction, and the authors do not use the selected layer to claim a new externally validated generalization. The downsampling of all input audio to 16 kHz (Section III-A: 'All audios are preprocessed ... and downsampled to 16kHz for compatibility with the pre-trained models in use') is a genuine limitation for the paper's 'limited bias regarding the sampling rate' interpretation, because high-frequency content above 8 kHz is never seen by the model; but that is an external-validity concern about whether the claim matches the method, not a circular derivation of the result from its own inputs. The one self-citation (SOMOS, [25]) appears only as future work and is not load-bearing. Therefore the paper's prediction chain is self-contained with respect to the challenge data and no circular step rises to the threshold defined here.

Assumptions & free parameters 7 free parameters · 4 assumptions · 0 invented entities

The central results depend on a set of hand-chosen hyperparameters and preprocessing choices, most notably the downsampling of all audio to 16 kHz, which is simultaneously a free parameter and a domain assumption about what information matters for MOS. The model itself does not introduce new physical entities, and the evaluation is grounded in external human labels, so the circularity burden is low. The free parameters listed here are choices the paper makes that materially affect the reported numbers.

free parameters (7)
  • Number of RBF centers = 16
    MOS targets are encoded as 16-dimensional vectors with centers evenly spaced in [1,5]; the resolution is chosen by hand and no sensitivity analysis is given.
  • RBF kernel width sigma = not reported
    Equation 1 defines f(x,xc)=exp(-||x-xc||^2/sigma^2) but the paper never states sigma's value, leaving the target encoding under-specified.
  • WavLM layer index = 6 for T16, 9 for best ablation
    Layer selection is empirical: Section III-D shows layer 9 outperforms layer 6, and picking the best layer post hoc adds a selection degree of freedom.
  • Target rating statistic = median for T16, mean for ablations
    Section III-A says T16 is trained on median ratings while ablations use mean ratings, which the paper notes improve performance; this choice affects all reported numbers.
  • Input downsampling rate = 16 kHz
    All 24/48 kHz audio is resampled to 16 kHz; this is a preprocessing choice that defines what the model can perceive and is central to the sampling-rate claim.
  • Number of Mamba blocks = not stated
    The architecture section says 'a series of Mamba-based blocks' but gives no count, so the capacity of the model is not fully reproducible.
  • Rating noise for label jitter = 0.0001
    Noise is added to rating values before RBF encoding to improve robustness; the magnitude is stated but its effect is not analyzed.
assumptions (4)
  • domain assumption Frozen SSL embeddings (WavLM-Large, Whisper, VGGish) capture the information needed for MOS prediction without finetuning.
    The method relies entirely on offline extracted features from pre-trained models; no SSL weights are updated during training.
  • domain assumption Downsampling to 16 kHz does not remove the acoustic cues that determine the MOS labels in Track 3.
    Section III-A applies sv56 normalization and resampling to all files; the validity of the cross-sampling-rate claim depends on this.
  • domain assumption The challenge's human MOS ratings and SRCC metric are a reliable ground truth.
    Performance is measured against ratings from 10 listeners per test and the official SRCC; the paper does not question label reliability.
  • domain assumption Mamba-2 selective state space layers can model the temporal dependencies in SSL feature sequences relevant to quality.
    The architecture adopts Mamba as the sequence encoder without a comparison against alternative sequence models like transformers in the same setup.

how reviews work

0 comments
Cite this review

Pith. "Pith review of MambaRate: Speech Quality Assessment Across Different Sampling Rates." pith.science (2026). https://pith.science/paper/Z7CBWIZ5

@misc{pith2026250712090,
  author       = {Pith},
  title        = {Pith review of: MambaRate: Speech Quality Assessment Across Different Sampling Rates},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/Z7CBWIZ5}},
  note         = {Machine review of arXiv:2507.12090}
}
read the original abstract

We propose MambaRate, which predicts Mean Opinion Scores (MOS) with limited bias regarding the sampling rate of the waveform under evaluation. It is designed for Track 3 of the AudioMOS Challenge 2025, which focuses on predicting MOS for speech in high sampling frequencies. Our model leverages self-supervised embeddings and selective state space modeling. The target ratings are encoded in a continuous representation via Gaussian radial basis functions (RBF). The results of the challenge were based on the system-level Spearman's Rank Correllation Coefficient (SRCC) metric. An initial MambaRate version (T16 system) outperformed the pre-trained baseline (B03) by ~14% in a few-shot setting without pre-training. T16 ranked fourth out of five in the challenge, differing by ~6% from the winning system. We present additional results on the BVCC dataset as well as ablations with different representations as input, which outperform the initial T16 version.

Figures

Figures reproduced from arXiv: 2507.12090 by the authors.

Figure 1
Figure 1. MambaRate’s architecture II. METHODOLOGY We present MambaRate, which is a lightweight solution that generalizes across different sampling rates. There is no requirement for any type of metadata, such as listener or system identifiers. The model utilizes only SSL features, without the need for finetuning a large pre-trained SSL model. On the contrary, the baseline includes the parameters of the SSL model during its t… view at source ↗
Figure 2
Figure 2. The initial version of MambaRate (T16) in the AudioMOS 2025 [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

27 extracted references · 15 canonical work pages

  1. [1]

    Mosnet: Deep learning-based objective assessment for voice conversion,

    C.-C. Lo, S.-W. Fu, W.-C. Huang, X. Wang, J. Yamagishi, Y . Tsao, and H.-M. Wang, “Mosnet: Deep learning-based objective assessment for voice conversion,” in Interspeech 2019, 2019, pp. 1541–1545, DOI: https://doi.org/10.21437/Interspeech.2019-2003

    doi:10.21437/interspeech.2019-2003 2019

  2. [3]

    Deepmos- B: Deep posterior mean-opinion- score using beta distribution,

    X. Liang, F. Cumlin, V . Ungureanu, C. K. A. Reddy, C. Sch ¨uldt, and S. Chatterjee, “Deepmos- B: Deep posterior mean-opinion- score using beta distribution,” in 2024 32nd European Sig- nal Processing Conference (EUSIPCO) , 2024, pp. 416–420, DOI: https://doi.org/10.23919/EUSIPCO63174.2024.10715351

    arXiv 2024

  3. [4]

    A review on subjective and objective evaluation of synthetic speech,

    E. Cooper, W.-C. Huang, Y . Tsao, H.-M. Wang, T. Toda, and J. Ya- magishi, “A review on subjective and objective evaluation of synthetic speech,” Acoustical Science and Technology, vol. 45, no. 4, pp. 161–183, 2024, DOI: https://doi.org/10.1250/ast.e24.12

    doi:10.1250/ast.e24.12 2024

  4. [5]

    A comparison of deep learn- ing mos predictors for speech synthesis quality,

    A. Ragano, E. Benetos, M. Chinen, H. B. Martinez, C. K. A. Reddy, J. Skoglund, and A. Hines, “A comparison of deep learn- ing mos predictors for speech synthesis quality,” in 2023 34th Irish Signals and Systems Conference (ISSC) , 2023, pp. 1–6, DOI: https://doi.org/10.1109/ISSC59246.2023.10162088

    work page arXiv 2023

  5. [6]

    wav2vec 2.0: a frame- work for self-supervised learning of speech representations,

    A. Baevski, H. Zhou, A. Mohamed, and M. Auli, “wav2vec 2.0: a frame- work for self-supervised learning of speech representations,” in Proceed- ings of the 34th International Conference on Neural Information Pro- cessing Systems, ser. NIPS ’20. Red Hook, NY , USA: Curran Associates Inc., 2020, DOI: https://dl.acm.org/doi/abs/10.5555/3495724.3496768

    arXiv 2020

  6. [7]

    Hubert: Self-supervised speech representation learn- ing by masked prediction of hidden units,

    W.-N. Hsu, B. Bolte, Y .-H. H. Tsai, K. Lakhotia, R. Salakhutdinov, and A. Mohamed, “Hubert: Self-supervised speech representation learn- ing by masked prediction of hidden units,” IEEE/ACM Trans. Audio, Speech and Lang. Proc. , vol. 29, p. 3451–3460, Oct. 2021, DOI: https://doi.org/10.1109/TASLP.2021.3122291

    arXiv 2021

  7. [8]

    Wavlm: Large-scale self-supervised pre- training for full stack speech processing,

    S. Chen, C. Wang, Z. Chen, Y . Wu, S. Liu, Z. Chen, J. Li, N. Kanda, T. Yoshioka, X. Xiao, J. Wu, L. Zhou, S. Ren, Y . Qian, Y . Qian, J. Wu, M. Zeng, X. Yu, and F. Wei, “Wavlm: Large-scale self-supervised pre- training for full stack speech processing,” IEEE Journal of Selected Topics in Signal Processing , vol. 16, no. 6, pp. 1505–1518, 2022, DOI: https...

    arXiv 2022

  8. [9]

    Mbnet: Mos prediction for synthesized speech with mean-bias network,

    Y . Leng, X. Tan, S. Zhao, F. Soong, X.-Y . Li, and T. Qin, “Mbnet: Mos prediction for synthesized speech with mean-bias network,” in ICASSP 2021 - 2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , 2021, pp. 391–395, DOI: https://doi.org/10.1109/ICASSP39728.2021.9413877

    arXiv 2021

Show all 27 references
  1. [10]

    Fusion of self-supervised learned models for mos prediction,

    Z. Yang, W. Zhou, C. Chu, S. Li, R. Dabre, R. Rubino, and Y . Zhao, “Fusion of self-supervised learned models for mos prediction,” in Interspeech 2022 , 2022, pp. 5443–5447, DOI: https://doi.org/10.21437/Interspeech.2022-10262

    2022 doi

  2. [11]

    Salf-mos: Speaker agnostic latent features downsampled for mos prediction,

    S. Agrawal, R. Gohil, G. K. Agrawal, V . C M, and K. Verma, “Salf-mos: Speaker agnostic latent features downsampled for mos prediction,” in 2024 International Conference on Signal Pro- cessing and Communications (SPCOM) , 2024, pp. 1–5, DOI: https://doi.org/10.1109/SPCOM60851....

    2024

  3. [12]

    The voicemos challenge 2024: Beyond speech quality prediction,

    W.-C. Huang, S.-W. Fu, E. Cooper, R. E. Zezario, T. Toda, H.- M. Wang, J. Yamagishi, and Y . Tsao, “The voicemos challenge 2024: Beyond speech quality prediction,” 2024. [Online]. Available: https://arxiv.org/abs/2409.07001

    2024 arXiv

  4. [13]

    Generalization ability of mos prediction networks,

    E. Cooper, W.-C. Huang, T. Toda, and J. Yamagishi, “Generalization ability of mos prediction networks,” in ICASSP 2022 - 2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2022, pp. 8442–8446

    2022

  5. [14]

    How do voices from past speech synthesis challenges compare today?

    E. Cooper and J. Yamagishi, “How do voices from past speech synthesis challenges compare today?” in 11th ISCA Speech Synthesis Workshop (SSW 11) , 2021, pp. 183–188

    2021

  6. [15]

    Transformers: State-of-the-art natural language processing,

    T. Wolf, L. Debut, V . Sanh, J. Chaumond, C. Delangue, A. Moi, P. Cistac, T. Rault, R. Louf, M. Funtowicz, J. Davison, S. Shleifer, P. von Platen, C. Ma, Y . Jernite, J. Plu, C. Xu, T. L. Scao, S. Gugger, M. Drame, Q. Lhoest, and A. M. Rush, “Transformers: State-of-the-art nat...

    2020

  7. [16]

    Towards out-of-distribution detection in vocoder recognition via latent feature reconstruction,

    R. Du, J. Yao, Q. Kong, and Y . Cao, “Towards out-of-distribution detection in vocoder recognition via latent feature reconstruction,”

  8. [17]

    Mish: A self regularized non-monotonic activation function,

    D. Misra, “Mish: A self regularized non-monotonic activation function,” in British Machine Vision Conference (BMVC) , 2020

    2020

  9. [18]

    Mamba4rec: Towards efficient sequential recommendation with selective state space models,

    C. Liu, J. Lin, J. Wang, H. Liu, and J. Caverlee, “Mamba4rec: Towards efficient sequential recommendation with selective state space models,”

  10. [19]

    Transformers are SSMs: Generalized models and efficient algorithms through structured state space duality,

    T. Dao and A. Gu, “Transformers are SSMs: Generalized models and efficient algorithms through structured state space duality,” in Interna- tional Conference on Machine Learning (ICML) , 2024

    2024

  11. [20]

    Available: https://arxiv.org/abs/2403.03900

    [Online]. Available: https://arxiv.org/abs/2403.03900

    arXiv

  12. [21]

    Predicting locations of cryptic pockets from single protein structures using the pocketminer graph neural network,

    A. Meller, M. Ward, J. Borowsky, M. Kshirsagar, J. M. Lotthammer, F. Oviedo, J. L. Ferres, and G. R. Bowman, “Predicting locations of cryptic pockets from single protein structures using the pocketminer graph neural network,” Nature Communications, vol. 14, no. 1, p. 1177, 2023

    2023

  13. [22]

    Fixing weight decay regularization in adam,

    I. Loshchilov and F. Hutter, “Fixing weight decay regularization in adam,” CoRR, vol. abs/1711.05101, 2017. [Online]. Available: http://arxiv.org/abs/1711.05101

    2017 arXiv

  14. [23]

    Cnn architectures for large-scale audio classification,

    S. Hershey, S. Chaudhuri, D. P. W. Ellis, J. F. Gemmeke, A. Jansen, R. C. Moore, M. Plakal, D. Platt, R. A. Saurous, B. Seybold, M. Slaney, R. J. Weiss, and K. Wilson, “Cnn architectures for large-scale audio classification,” in 2017 IEEE International Conference on Acoustics,...

    2017

  15. [24]

    Robust speech recognition via large-scale weak supervi- sion,

    A. Radford, J. W. Kim, T. Xu, G. Brockman, C. McLeavey, and I. Sutskever, “Robust speech recognition via large-scale weak supervi- sion,” in Proceedings of the 40th International Conference on Machine Learning, ser. ICML’23. JMLR.org, 2023

    2023

  16. [25]

    Somos: The Samsung open mos dataset for the evaluation of neural text-to-speech synthesis,

    G. Maniati, A. Vioni, N. Ellinas, K. Nikitaras, K. Klapsas, J. S. Sung, G. Jho, A. Chalamandaris, and P. Tsiakoulis, “Somos: The Samsung open mos dataset for the evaluation of neural text-to-speech synthesis,” in Interspeech 2022 , 2022, pp. 2388–2392

    2022

  17. [26]

    TensorFlow Model Garden,

    H. Yu, C. Chen, X. Du, Y . Li, A. Rashwan, L. Hou, P. Jin, F. Yang, F. Liu, J. Kim, and J. Li, “TensorFlow Model Garden,” https://github.com/tensorflow/models, 2020

    2020

  18. [28]

    The voice conversion challenge 2018: Promoting development of parallel and nonparallel methods,

    J. Lorenzo-Trueba, J. Yamagishi, T. Toda, D. Saito, F. Villavicencio, T. Kinnunen, and Z. Ling, “The voice conversion challenge 2018: Promoting development of parallel and nonparallel methods,” in The Speaker and Language Recognition Workshop (Odyssey 2018) , 2018, pp. 195–202

    2018

  19. [2024]

    Available: https://arxiv.org/abs/2406.02233

    [Online]. Available: https://arxiv.org/abs/2406.02233

    arXiv

Pith tools

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