Pith. sign in

REVIEW 4 major objections 6 minor 33 references

Dysarthria Normalization via Local Lie Group Transformations for Robust ASR

T0 review · 4 major / 6 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read Dysarthric speech can be normalized before recognition by learning local Lie-group warp fields from synthetically warped healthy spectrograms, cutting hard-case word error by up to 17 points.

desk verdict Promising idea with a real calibration flaw: the zero-shot claim depends on per-dataset epsilon tuning, but the effect is likely real and worth referee time. read the letter →

arxiv 2504.12279 v2 pith:QY32WRSP submitted 2025-04-16 cs.SD cs.CLcs.LGeess.AS

Mikhail Osipov This is my paper · ORCID
classification cs.SDcs.CLcs.LGeess.AS
keywords dysarthricspeechrecognitionLiegrouptransformationsspectrogramnormalizationzero-shotrobustnessspontaneoussymmetrybreakingworderrorratefront-endsyntheticdataaugmentation
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper argues that dysarthric speech is best understood as healthy speech passed through smooth, local time-frequency distortions, and that undoing those distortions is enough to make a generic ASR system work on it. The authors model each distortion as a local Lie group action on the spectrogram, covering time warps, frequency warps, two-dimensional warps, and amplitude changes, and train a network using only healthy audio that has been synthetically warped in this way. At test time the front-end is applied zero-shot to real dysarthric speech from the TORGO and UA-Speech corpora, improving word, character, and phoneme error rates, with the largest gains on exactly the utterances where the baseline ASR fails hardest, while clean speech is left unharmed. The proposal matters because it promises a robust, interpretable, and data-efficient route to disordered-speech recognition without paired clinical recordings.

What carries the argument

The carrier of the argument is the exponential map of an infinite-dimensional Lie algebra acting pointwise on the spectrogram. A generator of the form $X = (v+u_t)\partial_t + (w+u_f)\partial_f + \alpha + i\beta$ is integrated as $\tilde{S} = \exp(\epsilon X)[S]$, where the scalar fields $(v,w,u_t,u_f,\alpha)$ are predicted by a U-Net from the input magnitude spectrogram; the approximate inverse reverses the warps and divides out the amplitude field. A spontaneous-symmetry-breaking `hat` loss term keeps the predicted fields from collapsing to the trivial zero solution, and a curriculum increases the distortion scale $\epsilon$ from near-identity to strong warps during training. This machinery converts the geometric hypothesis into a differentiable, trainable front-end.

What would settle it

Run the trained front-end on held-out dysarthric speakers using the paper's fixed $\epsilon$ dictionary (for example, 0.6 for the RNNT backend) and compare with per-speaker tuned $\epsilon$ values. If the reported zero-shot gains disappear or become negative unless $\epsilon$ is re-tuned per speaker, the improvement is calibration rather than learned geometry.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that the varied distortions of dysarthric articulation can be approximated as local Lie group transformations of a clean spectrogram, and that a network trained exclusively on synthetic instances of those transformations learns an approximate inverse that transfers to real pathology it has never seen. Concretely, the system predicts dense scalar fields for time stretching, frequency stretching, 2D warping, and amplitude modulation; exponentiating these fields yields a correction, and the corrected spectrograms reduce WER on severe TORGO utterances by up to 17 percentage points, cut WER variance by 16 percent, and improve phoneme and character error rates in parallel, with no degradation on clean CommonVoice. The same pattern holds under two different ASR backends, indicating that the front-end itself, rather than a particular recognizer, is responsible for the gains.

Load-bearing premise

The zero-shot transfer rests on a calibration shortcut: the model predicts the shape of the correction, but the global strength of each correction is chosen by hand per dataset and per ASR backend, so a speaker whose distortion scale is not in that dictionary may be under- or over-corrected.

Editorial extensions

If this is right

  • A front-end trained purely on synthetic healthy-speech distortions can improve real dysarthric ASR without any pathological training data, so new speakers and severities do not require paired clinical recordings.
  • Hard cases benefit most: utterances with baseline WER above 50 percent show the largest absolute gains, and WER variance across the test set drops, meaning the recognizer's worst failures are the ones being repaired.
  • Because corrections reduce phoneme and character errors as well as word errors, the effect is linguistic rather than a re-scoring or hallucination artifact.
  • The improvement reproduces across two ASR backends, suggesting the front-end is architecture-independent rather than tuned to one recognizer.
  • Randomly generated but equally smooth Lie-style warps give only part of the benefit, so the input-conditioned content of the predicted fields is doing real work.

Reading between the lines

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

  • If synthetic-to-real transfer is this strong, the same field generator could be run in reverse to synthesize dysarthria-like distortions from any healthy corpus, turning the approach into a data-augmentation engine for ASR beyond the two backends tested here.
  • The per-dataset epsilon dictionary marks the natural next variable to learn: an auxiliary estimator of global distortion scale per utterance would make the method fully automatic and would directly test how much of the gain is geometry versus calibration.
  • Because the model ignores phase yet still reduces phoneme errors, amplitude and time-frequency geometry may carry most of the dysarthria signal; extending the framework to complex spectrograms would quantify the remaining contribution of phase distortions.
  • Treating the predicted fields as realizations of random fields invites prior-based regularization, such as articulatory smoothness or energy constraints, which could make corrections speaker-specific without paired data.
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 a geometry-driven front-end for dysarthric speech recognition. Speech distortions are modeled as local Lie group transformations applied to magnitude spectrograms, parameterized by scalar fields. A U-Net is trained on synthetically warped healthy CommonVoice speech to predict these fields, and an approximate inverse transformation is applied at test time to normalize pathological speech. Evaluation on TORGO, UA-Speech, and CommonVoice with two ASR backends (RNNT-Conformer and CrisperWhisper) shows WER reductions, particularly on high-WER subsets, and reports no degradation on clean speech. The central claim is that the learned fields provide zero-shot robustness gains on real dysarthric speech.

Significance. If the central claim holds, the paper introduces a novel, interpretable front-end that is trained entirely on synthetic data and generalizes to real pathological speech. The use of multiple ASR backends, an external random-warp baseline, and public datasets is a strength. However, the significance is currently limited by two issues: the global distortion scale is tuned per dataset and per backend at test time, which weakens the zero-shot claim, and the comparison baselines do not include established robust-ASR methods. With these addressed, the contribution could be valuable for the dysarthric ASR community.

major comments (4)
  1. [Evaluation Note, after Table II; Appendix G] The zero-shot claim is not yet established because the global scale of the predicted fields is chosen per dataset and per ASR backend at inference time. As stated in the Evaluation Note, RNNT-TORGO uses epsilon=0.6, RNNT-UA-Speech uses 2.0, RNNT-CommonVoice uses 0.5, and Whisper-All uses 3.0. Appendix G shows that predicted fields are normalized to [-1,1] and then rescaled by this externally supplied epsilon, so the entire strength of the correction is a hand-tuned parameter with access to the target evaluation condition. Please either provide a principled method for setting epsilon without test-set access (e.g., predicting it from the input), or report results across a range of epsilon values for each dataset and backend, and show that the reported gains are not dependent on a carefully chosen epsilon.
  2. [Section IV.F and Appendix J, Table VI] The random smooth-warp baseline does not control for the effect of choosing a favorable global scale. Table VI evaluates random transformations only on TORGO and only with epsilon=0.6, which is the value used for the learned model on that dataset. If different epsilon values had been used for the random baseline, its performance could change substantially, and the margin of the learned fields over random warps could shrink. Please sweep epsilon for the random baseline across a comparable range and compare against the learned model at the same epsilon values, or otherwise demonstrate that the learned input-conditioned fields outperform random fields at their respective optimal scales.
  3. [Section IV.E, Tables I and II] The paper claims that all improvements are statistically significant with p-values much smaller than 0.05, but no exact p-values, test statistics, or confidence intervals are reported. The tables give only point estimates of WER, PER, and variance, without error bars or across-run variability. This makes it difficult to judge the size and reliability of the reported gains, especially for differences of a few WER points (e.g., UA-Speech 116.5 to 110.8). Please report exact p-values, confidence intervals, and ideally results over multiple training seeds or evaluation subsets.
  4. [Section II, Related Work; Section IV.F, Ablations] The evaluation does not include established strong baselines for dysarthric ASR robustness, such as SpecAugment-based augmentation, VTLN, or voice-conversion normalization methods. The current comparisons are against identity and random smooth warps only. Without a comparison to a standard augmentation or normalization technique, the practical significance of the proposed Lie-group approach relative to much simpler methods is unclear. Please add at least one such baseline, or clearly motivate why these baselines are not appropriate for the zero-shot setting studied here.
minor comments (6)
  1. [Section II, Related Work] There is a typo in 'V ocal tract length normalization' (should be 'Vocal'), and the same spacing issue appears in 'V ocal tract deformations' in Section III-C. Please correct these.
  2. [Appendix J] The text says 'These results are reported in Table VI of the Appendix', but Table VI appears in the main text after Table V, and the numbering is inconsistent with the in-text references. Please renumber the tables and ensure all references point to the correct locations.
  3. [Section III-H and Appendix G] The use of the symbol epsilon is overloaded: it denotes the transformation strength in the training curriculum, the per-field scaling dictionary, and the inference-time scale in Appendix G. Please use distinct notation for these three concepts to avoid confusion.
  4. [Appendix L, Figure 6] The figure caption shows the original transcription as 'peace', the transformed transcription as 'pace', and the true label as 'paste'. This example may not illustrate an improvement, since 'pace' is not closer to 'paste' than 'peace' is. Please explain the intended interpretation or choose a clearer example.
  5. [Section IV-B] The U-Net uses a ResNeXt-50 encoder pretrained on ImageNet. Please clarify whether the input magnitude spectrogram is normalized as an image and whether the pretrained weights are used with the same channel count, since RGB ImageNet weights are typically 3-channel.
  6. [Section III-C] The monotonicity constraint tau'(t) > 0 is stated but it is not described how it is enforced in the discretized warping implementation. Please state whether the predicted time-warp fields are constrained during inference or only during training.

Circularity Check

1 steps flagged · score 2.0 of 10

Zero-shot WER gains depend on a test-set-calibrated epsilon dictionary rather than purely predicted fields; no other circularity found.

  1. fitted input called prediction [Section IV-E (Evaluation Note after Table II); Section III-H; Appendix G (rescaling fields)]
    "To accommodate different ASR backends and dataset characteristics, we used varying transformation strength values (denoted as ε) during inference. Specifically: RNNT–Torgo: ε = 0.6; RNNT–UASpeech: ε = 2.0; RNNT–CommonVoice: ε = 0.5; Whisper (CrisperWhisper)–All: ε = 3.0. These values were chosen empirically to balance transformation strength with ASR performance across conditions."

    Appendix G makes the mechanism explicit: ϕ_real = ϕ_pred · ε_type, after fields are normalized to [-1,1]. Thus the network predicts only the spatial shape of the deformation; the global severity of the correction is a constant chosen per evaluation corpus and per ASR backend with access to the target data. The reported zero-shot gains (including the headline TORGO>50 WER drop from 110.2 to 92.4) are therefore not fully predicted quantities: they incorporate a test-set-fitted scale. Because the random-warp baseline in Table VI is run only with ε=0.6 on TORGO, it does not control for favorable-scale selection; the learned-vs-random comparison remains meaningful at that fixed ε, but the absolute zero-shot claim is partly constructed by the externally supplied dictionary.

full rationale

The paper's main derivation is not circular: the U-Net is trained on synthetic spectrogram distortions with known generator fields from CommonVoice, never on TORGO or UA-Speech, and evaluated against external ASR backends; there is no self-citation chain, and no equation is defined in terms of the result it claims to derive. The only load-bearing evaluation leak is the per-dataset, per-backend ε dictionary at inference, which supplies a scale the model does not predict. This injects target-condition information into the zero-shot evaluation, but because the core claim of learned-field superiority over random smooth warps is tested at a fixed ε (0.6), the result still has independent content. Score 2 reflects a minor, disclosed, non-derivational circularity rather than a forced prediction.

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

The method introduces scalar transformation fields and a symmetry-breaking loss as modeling devices, not new physical entities. The central claim rests on the synthetic-to-real transfer assumption and on per-dataset scaling parameters tuned at inference.

free parameters (4)
  • Inference scale epsilon per dataset/backend = RNNT-TORGO 0.6; RNNT-UA-Speech 2.0; RNNT-CommonVoice 0.5; Whisper-all 3.0
    Chosen empirically on the evaluation sets to balance transformation strength and ASR performance (Section IV-E Evaluation Note). The model only predicts normalized fields; actual magnitude comes from this tuned scalar.
  • Loss weights lambda = Not fully specified in the paper
    Six loss terms (spec, cosine, kinetic, ssb, sparse) have tunable coefficients; the final values are not listed in Section IV-C.
  • Synthetic field generator parameters = Not fully specified (blob count, mask radius, amplitudes)
    The distribution of synthetic warps is controlled by these parameters (Appendix E); no exact values or ranges are given.
  • Curriculum schedule parameters = epsilon from 0.1 to 4.0, linear with warmup
    The schedule itself (warmup, plateau, final increase) is a free choice made for training stability (Section IV-D).
assumptions (5)
  • domain assumption Dysarthric speech distortions can be modeled as smooth, local, invertible Lie group transformations of the magnitude spectrogram, excluding clipping, blur, and dropped phonemes.
    This is the paper's core modeling assumption (Section III, III-B). If real dysarthria contains non-invertible or non-smooth distortions, the inverse will not capture them.
  • ad hoc to paper Synthetically generated smooth sinusoidal blob fields resemble real dysarthric distortion statistics sufficiently for zero-shot transfer.
    Training uses only synthetic warps (Section III-F, Appendix E); no real disordered data is used to validate the match.
  • domain assumption First-order approximation of the exponential map is sufficient for the distortions encountered in training and evaluation.
    The inverse is implemented with negated displacements and safe division (Section III-H, Appendix C); large distortions may need higher-order terms.
  • domain assumption Phase information can be discarded without losing information relevant to ASR normalization.
    The model acts on magnitude only (Section III-E); pitch and voicing distortions in phase are ignored.
  • ad hoc to paper The per-dataset and per-backend epsilon values tuned on the test sets remain valid for new speakers and conditions.
    Evaluation uses fixed epsilon per condition (Evaluation Note); any shift in distortion severity would require re-tuning.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Dysarthria Normalization via Local Lie Group Transformations for Robust ASR." pith.science (2026). https://pith.science/paper/QY32WRSP

@misc{pith2026250412279,
  author       = {Pith},
  title        = {Pith review of: Dysarthria Normalization via Local Lie Group Transformations for Robust ASR},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QY32WRSP}},
  note         = {Machine review of arXiv:2504.12279}
}
read the original abstract

We present a geometry-driven method for normalizing dysarthric speech by modeling time, frequency, and amplitude distortions as smooth, local Lie group transformations of spectrograms. Scalar fields generate these deformations via exponential maps, and a neural network is trained - using only synthetically warped healthy speech - to infer the fields and apply an approximate inverse at test time. We introduce a spontaneous-symmetry-breaking (SSB) potential that encourages the model to discover non-trivial field configurations. On real pathological speech, the system delivers consistent gains: up to 17 percentage-point WER reduction on challenging TORGO utterances and a 16 percent drop in WER variance, with no degradation on clean CommonVoice data. Character and phoneme error rates improve in parallel, confirming linguistic relevance. Our results demonstrate that geometrically structured warping provides consistent, zero-shot robustness gains for dysarthric ASR.

Figures

Figures reproduced from arXiv: 2504.12279 by the authors.

Figure 1
Figure 1. Examples of generated fields: ϕtime, ϕfreq (before masking), ϕamp (masked) G. Applying Spectrogram Transformations for Training To generate training data, we apply randomly sampled smooth transformation fields to clean spectrograms from the CommonVoice [4] dataset. These fields are drawn from a para￾metric Lie algebra (see Appendix E) and represent synthetic dysarthria-like distortions. Each transformation is applie… view at source ↗
Figure 3
Figure 3. Scaled loss function terms and ε dynamics across training steps. Model version v.2 selected CrisperWhisper as an additional ASR backend due to its training objective: unlike standard Whisper models, it is optimized for verbatim transcription, reducing paraphras￾ing and hallucinations. This makes it especially suitable for dysarthric datasets, which consist of short, clearly segmented utterances where literal transcr… view at source ↗
Figure 2
Figure 2. Training and validation loss dynamics (10000 samples from Common [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Training and validation loss dynamics (10000 samples from Common [PITH_FULL_IMAGE:figures/full_fig_p013_4.png]
Figure 5
Figure 5. Figure 5: Weighted loss function terms and ε dynamics across training steps. Model version v.1 suggesting its capacity to selectively normalize difficult speech segments without degrading performance on cleaner examples. TABLE III WER AND CER ACROSS DIFFERENT DATASETS AND SPECTR…
Figure 6
Figure 6. Figure 6: Example of input spectrogram (first picture from above) and predicted [PITH_FULL_IMAGE:figures/full_fig_p014_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

33 extracted references · 22 canonical work pages

  1. [1]

    Dysarthric speech databases for universal access research,

    H.-Y . Kim, M. Hasegawa-Johnson, M. Perlman, J. Gunderson, S.-E. Huang, K. Watkin, and E. Wealthy, “Dysarthric speech databases for universal access research,” Interspeech, pp. 1741–1744, 2008

    work page 2008

  2. [2]

    Torgo database of dysarthric articulation,

    F. Rudzicz, A. Namasivayam, and T. Wolff, “Torgo database of dysarthric articulation,” https://torgo.rrg.utoronto.ca/, 2012, accessed 2024-12-01

    work page 2012

  3. [3]

    Specaugment: A simple data augmentation method for automatic speech recognition,

    D. S. Park, W. Chan, Y . Zhang, C.-C. Chiu, B. Zoph, E. D. Cubuk, and Q. V . Le, “Specaugment: A simple data augmentation method for automatic speech recognition,” in Proc. Interspeech, 2019, pp. 2613– 2617

    work page 2019

  4. [4]

    Common voice: A massively-multilingual speech corpus,

    R. Ardila, M. Branson, K. Davis, M. Kohler, J. Meyer, R. Henretty, M. Morais, L. Saunders, F. Tyers, and G. Weber, “Common voice: A massively-multilingual speech corpus,” https://commonvoice.mozilla. org/en/datasets, 2020, version 17.0, accessed 2025-03-01

    work page 2020

  5. [5]

    Fastdiff: A fast conditional diffusion model for high-quality speech synthesis,

    Z.-H. Huang, Y .-A. Wu, X. Yang, L. Xie, and H.-y. Lee, “Fastdiff: A fast conditional diffusion model for high-quality speech synthesis,” in Proceedings of the AAAI Conference on Artificial Intelligence , vol. 36, no. 10, 2022, pp. 11 079–11 086

    work page 2022

  6. [6]

    Ffjord: Free-form continuous dynamics for scalable reversible generative models,

    W. Grathwohl, R. T. Chen, J. Bettencourt, I. Sutskever, and D. Duve- naud, “Ffjord: Free-form continuous dynamics for scalable reversible generative models,” arXiv preprint arXiv:1810.01367 , 2019

    arXiv 2019

  7. [7]

    Neural ordinary differential equations,

    R. T. Chen, Y . Rubanova, J. Bettencourt, and D. K. Duvenaud, “Neural ordinary differential equations,” in Advances in Neural Information Processing Systems, vol. 31, 2018

    work page 2018

  8. [8]

    Shapes and diffeomorphisms,

    L. Younes, “Shapes and diffeomorphisms,” 2019

    work page 2019

Show all 33 references
  1. [9]

    Geometric deep learning: Grids, groups, graphs, geodesics, and gauges,

    M. M. Bronstein, J. Bruna, T. Cohen, and P. Veliˇckovi´c, “Geometric deep learning: Grids, groups, graphs, geodesics, and gauges,” arXiv preprint arXiv:2104.13478, 2021

    2021 arXiv

  2. [10]

    Cyclegan-vc: Non-parallel voice conver- sion using cycle-consistent adversarial networks,

    T. Kaneko and H. Kameoka, “Cyclegan-vc: Non-parallel voice conver- sion using cycle-consistent adversarial networks,” in2018 26th European Signal Processing Conference (EUSIPCO) . IEEE, 2018, pp. 2100– 2104

    2018

  3. [11]

    Unidiffusion: A unified framework for arbitrary-domain voice conversion via diffusion models,

    Y . Li, X. Li, and K. Lee, “Unidiffusion: A unified framework for arbitrary-domain voice conversion via diffusion models,” arXiv preprint arXiv:2406.08568, 2024

    2024 arXiv

  4. [12]

    V oiceid- vc: V oice conversion with speaker-consistent latent representation and self-supervised voiceprint,

    X. Zhang, S. Yu, L. Liu, Y . Liu, Q. Wu, Y . Lin, and X. Wang, “V oiceid- vc: V oice conversion with speaker-consistent latent representation and self-supervised voiceprint,” arXiv preprint arXiv:2411.01710 , 2024

    2024 arXiv

  5. [13]

    Improving dysarthric speech intelligibility using cycle-consistent generative adversarial networks,

    Z.-H. Tan, W. Liu, Q. Wang, and H. Li, “Improving dysarthric speech intelligibility using cycle-consistent generative adversarial networks,” in ICASSP 2020 - IEEE International Conference on Acoustics, Speech and Signal Processing . IEEE, 2020, pp. 6274–6278

    2020

  6. [14]

    Representation of vocal tract length trans- formation based on group theory,

    A. Miyashita and T. Toda, “Representation of vocal tract length trans- formation based on group theory,” in ICASSP 2023 - 2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2023, pp. 1–5

    2023

  7. [15]

    Differentiable representation of warping based on lie group theory,

    ——, “Differentiable representation of warping based on lie group theory,” in 2023 IEEE Workshop on Applications of Signal Processing to Audio and Acoustics (WASPAA) , 2023, pp. 1–5

    2023

  8. [16]

    Group-structured deformations for data-driven signal modeling,

    H. Youn et al. , “Group-structured deformations for data-driven signal modeling,” arXiv preprint arXiv:2103.12345 , 2021

    2021 arXiv

  9. [17]

    Fant, Acoustic Theory of Speech Production: With Calculations Based on X-ray Studies of Russian Articulations

    G. Fant, Acoustic Theory of Speech Production: With Calculations Based on X-ray Studies of Russian Articulations . The Hague: Mouton, 1970

    1970

  10. [18]

    M. J. Ball and N. M ¨uller, Introduction to Clinical Phonetics and Linguistics. Wiley-Blackwell, 2004

    2004

  11. [19]

    Differential diagnostic patterns of dysarthria,

    F. L. Darley, A. E. Aronson, and J. R. Brown, “Differential diagnostic patterns of dysarthria,” Journal of Speech and Hearing Research, vol. 12, no. 2, pp. 246–269, 1969

    1969

  12. [20]

    Representation learning with contrastive predictive coding,

    A. van den Oord, Y . Li, and O. Vinyals, “Representation learning with contrastive predictive coding,” arXiv preprint arXiv:1807.03748 , 2018

    2018 arXiv

  13. [21]

    wav2vec 2.0: A framework for self-supervised learning of speech representations,

    A. Baevski, Y . Zhou, A. Mohamed, and M. Auli, “wav2vec 2.0: A framework for self-supervised learning of speech representations,” in Advances in Neural Information Processing Systems , vol. 33, 2020, pp. 12 449–12 460

    2020

  14. [22]

    Multi-view supervision for single-view reconstruction via differentiable ray consistency,

    S. Tulsiani, T. Zhou, A. A. Efros, and J. Malik, “Multi-view supervision for single-view reconstruction via differentiable ray consistency,” in CVPR, 2017

    2017

  15. [23]

    View synthesis by appearance flow,

    T. Zhou, S. Tulsiani, W. Sun, J. Malik, and A. A. Efros, “View synthesis by appearance flow,” in ECCV, 2016

    2016

  16. [24]

    Stochastic backpropa- gation and approximate inference in deep generative models,

    D. J. Rezende, S. Mohamed, and D. Wierstra, “Stochastic backpropa- gation and approximate inference in deep generative models,” in ICML, 2014, pp. 1278–1286

    2014

  17. [25]

    U-net: Convolutional networks for biomedical image segmentation,

    O. Ronneberger, P. Fischer, and T. Brox, “U-net: Convolutional networks for biomedical image segmentation,” in International Conference on Medical image computing and computer-assisted intervention. Springer, 2015, pp. 234–241

    2015

  18. [26]

    Mel-cepstral distance measure for objective speech qual- ity assessment,

    R. Kubichek, “Mel-cepstral distance measure for objective speech qual- ity assessment,” in IEEE Pacific Rim Conference on Communications Computers and Signal Processing . IEEE, 1993, pp. 125–128

    1993

  19. [27]

    An end-to-end model for asr error mini- mization in text-to-speech synthesis,

    S. Takaki and J. Yamagishi, “An end-to-end model for asr error mini- mization in text-to-speech synthesis,” in ICASSP. IEEE, 2021

    2021

  20. [28]

    Nemo: a toolkit for building ai applications using neural modules,

    O. Kuchaiev, B. Ginsburg, Y . Li, V . Lavrukhin, D. Park, J. Ernst, S. Kriman, V . Dwivedi, Y . Zhang, R. Leary, and et al., “Nemo: a toolkit for building ai applications using neural modules,” arXiv preprint arXiv:1909.09577, 2019

    1909 arXiv

  21. [29]

    Robust speech recognition via large-scale weak supervision,

    A. Radford, J. W. Kim, T. Xu, G. Brockman, C. McLeavey, and I. Sutskever, “Robust speech recognition via large-scale weak supervision,” 2022. [Online]. Available: https://arxiv.org/abs/2212.04356

    2022 arXiv

  22. [30]

    Crisperwhisper: Accurate timestamps on verbatim speech transcriptions,

    L. Wagner, B. Thallinger, and M. Zusag, “Crisperwhisper: Accurate timestamps on verbatim speech transcriptions,” 2024. [Online]. Available: https://arxiv.org/abs/2408.16589

    2024 arXiv

  23. [31]

    J. E. Marsden and T. S. Ratiu, Introduction to Mechanics and Symmetry: A Basic Exposition of Classical Mechanical Systems . Springer, 1999

    1999

  24. [32]

    Cohen, Time-Frequency Analysis

    L. Cohen, Time-Frequency Analysis. Prentice Hall, 1995

    1995

  25. [33]

    V . G. Kac, Infinite-dimensional Lie Algebras , 3rd ed. Cambridge University Press, 1994

    1994

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.