Pith. sign in

REVIEW 4 major objections 6 minor 34 references

Improved Intelligibility of Dysarthric Speech using Conditional Flow Matching

T0 review · 4 major / 6 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read Using WavLM-derived discrete acoustic units rather than mel-spectrograms, a conditional flow matching model converts dysarthric speech into intelligible single-voice clean speech, achieving WER 31.30 versus 84.07 and faster convergence.

desk verdict Plausible CFM+units pipeline for dysarthric conversion, but the units-vs-mel comparison is confounded by training budget and one table row contradicts the text. read the letter →

arxiv 2506.16127 v1 pith:BYWHW5QK submitted 2025-06-19 cs.SD cs.AIeess.AS

Shoutrik Das , Nishant Singh , Arjun Gangwar , S Umesh This is my paper · ORCID
classification cs.SDcs.AIeess.AS
keywords dysarthricspeechconditionalflowmatchingdiscreteacousticunitsWavLMself-supervisedlearningconversionintelligibilitydiffusiontransformer
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 seeks to establish that dysarthric speech can be converted into intelligible, natural-sounding clean speech without a paired parallel corpus, using a fully non-autoregressive Conditional Flow Matching (CFM) Diffusion Transformer. The key proposal is to feed the model discrete acoustic units extracted from WavLM rather than mel-spectrograms, on the theory that the units act as a speaker-invariant bottleneck that preserves phonetic content while discarding dysarthric distortions. On the Speech Accessibility Project test set, the units-based model reaches a word error rate of 31.30 and a mean opinion score of 3.9, while the same architecture fed mel-spectrograms reaches WER 84.07 and MOS 1.0, and converges much more slowly. The clean training targets are synthesized with a single-speaker text-to-speech prompt, so the output also comes out in one consistent voice.

What carries the argument

The central machinery is Conditional Flow Matching with a Diffusion Transformer, paired with discrete acoustic units as the input representation. CFM constructs a probability path between noise and data using an optimal-transport conditional flow, and the model learns the vector field $v_t(x;\theta)$ by regressing onto the conditional velocity $u_t(x|x_1)$; at inference an ODE solver integrates from noise to the target mel-spectrogram. The discrete units are WavLM-Large layer-21 features quantized to 512 K-means cluster IDs, temporally collapsed by merging consecutive repeats, and embedded as tokens; this is the speaker-invariant bottleneck that the paper argues carries phonetic content while removing speaker and prosodic variability. The same model is trained in an infilling setup on masked mel-spectrograms, following F5-TTS's inference strategy.

What would settle it

Compute the word error rate of the F5-TTS-synthesized clean targets themselves against the original transcripts; a target WER near 31.30 would show that the reported intelligibility gain is inherited from the TTS rather than produced by the conversion model.

Watch

Extended reading notes

Core claim

The paper's central claim is that replacing mel-spectrogram inputs with discrete acoustic units extracted from WavLM's 21st layer and quantized with a 512-cluster K-means model makes dysarthric-to-clean speech conversion dramatically more effective under a Conditional Flow Matching (CFM) Diffusion Transformer. CFM-base-units reaches WER 31.30 and MOS 3.9 on the SAPC test set, while CFM-base-MEL reaches WER 84.07 and MOS 1.0; units also reach intelligible output after roughly one million combined updates whereas the mel model does not after two million. The model maps collapsed unit sequences from dysarthric audio, together with a masked clean mel-spectrogram context, to the target mel-spectrogram, using F5-TTS with a single-speaker prompt to generate clean training targets and BigVGAN to vocode the output. This is presented as evidence that discrete acoustic units are a better intermediate representation than mel-spectrograms for dysarthric speech restoration.

Load-bearing premise

The load-bearing assumption is that the F5-TTS speech synthesized from the dataset transcripts is a valid clean ground truth for training; if the transcripts or the TTS introduce errors, the model is trained to reproduce those errors as if they were correct speech.

Editorial extensions

If this is right

  • If the central claim holds, discrete acoustic units are the right intermediate representation for dysarthric speech conversion, because they compress away speaker identity and align more naturally with clean targets than mel-spectrograms do.
  • The CFM approach provides a non-autoregressive alternative to GAN-based conversion, with straight-line probability paths that make training and ODE-based inference straightforward.
  • Because fine-tuning with one hour of target-speaker speech restores performance on unseen speakers, deployment for a new user only needs a small adaptation set.
  • The absence of a parallel corpus is filled by synthetic single-speaker TTS targets, so the same recipe can be applied to any dysarthric dataset that has transcripts.

Reading between the lines

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

  • The 31.30 WER may partly reflect the quality ceiling of the F5-TTS synthetic targets themselves; if a recognizer already transcribes those targets with a similar WER, the conversion model's contribution is narrower than the headline gap suggests.
  • Because discrete units are language-agnostic cluster IDs, the same pipeline could transfer to non-English dysarthric speech by retraining only the K-means quantizer and fine-tuning on a few hours of the target language.
  • A direct comparison of the CFM-units system against Unit-DSR on matched test speech would isolate whether the gain comes from WavLM's denoising-based training or from the flow-matching generator, since both approaches use discrete units.
  • Substituting a learned neural quantizer for K-means, or varying the number of clusters, is a natural knob to test whether sub-phonemic detail is being lost in the bottleneck.
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 dysarthric-to-clean speech conversion system based on Conditional Flow Matching (CFM) with Diffusion Transformers, using quantized WavLM discrete acoustic units as the intermediate representation instead of mel-spectrograms. The authors generate clean reference speech with F5-TTS from transcripts of the Speech Accessibility Project corpus, pretrain on LibriSpeech, and fine-tune on dysarthric speech. They report that the unit-based model achieves WER 31.30 and MOS 3.9, while a mel-spectrogram counterpart achieves WER 84.07 and MOS 1.0, and they claim faster convergence for units. They also report a seen/unseen speaker breakdown and an experiment with 1 hour of target-speaker fine-tuning.

Significance. If the claims hold, the paper offers a practical direction for dysarthric speech conversion: using SSL-derived discrete units as a speaker-invariant bottleneck inside a non-autoregressive flow-matching model, and generating a single-speaker clean target with a TTS model to avoid the need for a parallel corpus. The large reported WER gap between units and mel is potentially valuable and the setup is reproducible in principle. The strengths are the clear task framing, the use of WavLM for noise robustness, and the attempt to evaluate both ASR-based intelligibility and human MOS.

major comments (4)
  1. [Section 5, Table 1 and convergence discussion] The comparison between CFM-base-units and CFM-base-MEL is not controlled for training budget. The text states that the unit model achieved intelligible speech after 1M updates while the mel model failed even after 2M updates, but the paper does not report the total number of updates for CFM-base-MEL, does not provide learning curves for either model, and does not present a matched-update evaluation. As written, the 84.07 WER and 1.0 MOS for the mel model could reflect an undertrained or under-tuned baseline rather than an intrinsic limitation of mel input, so the central faster-convergence and intelligibility claims are not yet cleanly established.
  2. [Section 5, Table 2] The 'Unseen finetuned' row reports WER 30.90 but MOS 1.0 with confidence interval 1.00–1.00, which contradicts the text's claim that fine-tuning with just 1 hour of target-speaker speech overcomes the seen/unseen gap. If MOS was collected for this condition, the floor value is inconsistent with the improved WER; if MOS was not collected, the table should not contain a numeric entry. Please clarify this discrepancy, as the speaker-adaptation claim rests on this row.
  3. [Section 3.1] The training target is synthesized by F5-TTS from SAPC transcripts, so the model is trained to reproduce any ASR or TTS errors present in those transcripts. The paper does not discuss the accuracy of the synthetic targets or measure how often the transcripts match the dysarthric audio, yet the WER evaluation uses the same transcription source. This is a load-bearing validity concern because the system is not evaluated against a verified clean-speech reference, and the claim of conversion to 'regular' or 'clean' speech is therefore only relative to the synthesized targets.
  4. [Section 5 and Conclusion] The conclusion states that CFM serves as a viable alternative to GANs and traditional methods for dysarthric speech conversion, but the paper evaluates no GAN-based or prior DSC system, including Unit-DSR, which is discussed as the most closely related work. Adding at least one strong baseline (e.g., Unit-DSR or a GAN-based voice-conversion system) is necessary to support the comparative claim that the proposed approach is competitive or superior.
minor comments (6)
  1. [Section 2] There is a typo: 'inavailablity' should be 'unavailability'.
  2. [Section 3.4] There is a typo: 'accoustic units' should be 'acoustic units'.
  3. [Figure 1 caption] The caption says 'Diagram of the Dysarthric speech recognition system,' but the system performs speech conversion, not recognition; please correct the caption.
  4. [Table 1] The MOS values in Table 1 are reported without confidence intervals or the number of utterances per condition; with only 23 raters and 10 randomly sampled utterances, please provide CIs or raw scores to allow assessment of the MOS difference.
  5. [Section 3.2] The selection of the 21st WavLM layer and the K-Means cluster count of 512 is not justified; please provide a reference or an ablation/justification for these hyperparameters.
  6. [Section 5] The paper does not report the number of speakers in the test set or the distribution of dysarthria severities, which is needed to interpret the WER and MOS results and to assess generalization claims.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the paper's claims are empirical comparisons, not derivations, and no fitted parameter is relabeled as a prediction.

full rationale

The paper makes an empirical claim that discrete WavLM acoustic units outperform mel-spectrograms for dysarthric-to-clean speech conversion using a Conditional Flow Matching model. There is no derivation chain in which an output quantity is defined in terms of the quantity it is supposed to predict. No fitted parameter is renamed as a prediction: the K-means quantizer is trained on external corpora (LibriSpeech, GigaSpeech, VoxPopuli), the CFM objective is standard flow matching from Lipman et al., and the reported WER and MOS numbers are direct evaluations of generated audio. The paper does not invoke a uniqueness theorem or any load-bearing self-citation; its references to prior work such as Unit-DSR and F5-TTS are either baselines or component choices, not justifications that make the result true by construction. The reader's flagged concern that the F5-TTS-synthesized audio is used as the clean training target is a data-construction issue, not circular reasoning: the model is not defined in terms of its own output, and the evaluation does not presuppose the conclusion. The skeptic's concern that the mel baseline may have been undertrained (no convergence curves, no matched-update comparison) is a legitimate experimental-fairness or correctness risk, but it does not amount to circularity. Accordingly, the appropriate finding is no significant circularity, score 0.

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

The central claim rests on three categories of support: standard flow-matching mathematics from prior work, and two domain assumptions specific to this pipeline. The first domain assumption is that TTS-synthesized audio is a valid ground-truth target for dysarthric conversion. The second is that discrete units trained on normal speech remain reliable phonetic descriptors for disordered speech. Neither is independently verified in the paper. The hand-chosen K-Means size, WavLM layer, and VAD threshold are additional tuning decisions that affect the reported results.

free parameters (3)
  • K-Means cluster count (K) = 512
    The number of discrete acoustic units is chosen by hand (Section 3.2). The quantization granularity directly controls the information bottleneck in the input representation and is not derived from the data or an external benchmark.
  • WavLM layer index = 21
    Features are extracted from the 21st layer of WavLM-Large (Section 3.2). The layer choice affects what phonetic information the units capture and is selected by the authors without a comparative study.
  • VAD silence-removal threshold = not reported
    Section 3.4.2 specifies a 'higher speech-probability threshold' for leading and trailing silences without giving values. The choice affects the temporal alignment between dysarthric units and clean mel targets.
assumptions (3)
  • standard math CFM optimal-transport conditional paths give identical gradients to the marginal flow-matching objective (Lipman et al.).
    Invoked in Section 3.3 (Eqs. 1-5). The paper uses the OT path from prior work without reproving it.
  • domain assumption Synthesized F5-TTS speech is an adequate surrogate for 'clean' speech of the same content and preserves phonetic content.
    Section 3.1: clean targets are generated by F5-TTS from text with a single-speaker prompt, without verification that the generated audio is error-free or that prosody matches the intended utterance.
  • domain assumption WavLM features and K-Means centroids trained on typical speech remain phonetically discriminative when applied to dysarthric speech.
    Section 3.2 and 3.4: units are extracted from WavLM features with a K-Means model trained on LibriSpeech, GigaSpeech, and VoxPopuli. The paper does not evaluate unit error rates on dysarthric audio.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Improved Intelligibility of Dysarthric Speech using Conditional Flow Matching." pith.science (2026). https://pith.science/paper/BYWHW5QK

@misc{pith2026250616127,
  author       = {Pith},
  title        = {Pith review of: Improved Intelligibility of Dysarthric Speech using Conditional Flow Matching},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BYWHW5QK}},
  note         = {Machine review of arXiv:2506.16127}
}
read the original abstract

Dysarthria is a neurological disorder that significantly impairs speech intelligibility, often rendering affected individuals unable to communicate effectively. This necessitates the development of robust dysarthric-to-regular speech conversion techniques. In this work, we investigate the utility and limitations of self-supervised learning (SSL) features and their quantized representations as an alternative to mel-spectrograms for speech generation. Additionally, we explore methods to mitigate speaker variability by generating clean speech in a single-speaker voice using features extracted from WavLM. To this end, we propose a fully non-autoregressive approach that leverages Conditional Flow Matching (CFM) with Diffusion Transformers to learn a direct mapping from dysarthric to clean speech. Our findings highlight the effectiveness of discrete acoustic units in improving intelligibility while achieving faster convergence compared to traditional mel-spectrogram-based approaches.

Figures

Figures reproduced from arXiv: 2506.16127 by the authors.

Figure 1
Figure 1. Diagram of the Dysarthric speech recognition system A critical aspect of CFM is the choice of the conditional flow. Since a flow defines trajectories that transition from p0 to p1, simpler trajectories (e.g., straight-line paths) can enhance learning efficiency and improve IVP-solving accuracy. [27] proposed an optimal transport (OT) path, the time-dependent conditional probability path and corresponding vector fiel… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

34 extracted references · 19 canonical work pages

  1. [1]

    Dysarthria arises from neurological conditions that im- pair the control and coordination of the muscles involved in speech production

    Introduction Effective communication is fundamental to human interac- tion, yet individuals with dysarthria—a motor speech disor- der—encounter profound challenges in expressing themselves clearly. Dysarthria arises from neurological conditions that im- pair the control and coordination of the muscles involved in speech production. As a result, affected i...

    work page

  2. [2]

    Related work Converting Dysarthric speech to regular speech presents two primary challenges: the absence of parallel dysarthric-clean speech corpus and the complexity of learning a robust trans- formation function. The first major hurdle in developing dysarthric-to-regular speech conversion (DSC) systems is the inherent inavailabil- ity of a parallel dysa...

    work page Pith review arXiv 2025

  3. [3]

    We begin by describing our approach for generating reg- ular speech from dysarthric speech

    Method In this section, we outline the methodology employed in our work. We begin by describing our approach for generating reg- ular speech from dysarthric speech. Next, we introduce the quantization technique used to obtain discrete acoustic units and discuss their relationship to various speech characteristics. Fol- lowing this, we present our proposed...

    work page

  4. [4]

    Training setup All speech signals were resampled to 16 kHz during both fea- ture extraction with WavLM and the training process. For the target representation, mel spectrograms were generated using a window length of 40 ms, hop length of 10 ms, and an FFT size of 1024 points, with 80 mel channels to maintain com- patibility with our vocoder, BigVGAN, trai...

    work page

  5. [5]

    As baselines, we use wav2vec2-base-960h [29] and its finetuned counterpart trained on the same SAPC dataset

    Experimental Results We evaluate our model on a subset of the Speech Accessibil- ity Project dataset reserved for testing, which includes speech from both seen and unseen speakers. As baselines, we use wav2vec2-base-960h [29] and its finetuned counterpart trained on the same SAPC dataset. We assess transcription quality by evaluating wav2vec2 (without fin...

    work page

  6. [6]

    Our experiments show that acoustic discrete units not only achieve significantly faster convergence than mel-spectrograms but also produce speech with higher intelligibility

    Conclusion We propose the use of acoustic discrete units as an alterna- tive to mel-spectrograms for dysarthric-to-clean speech genera- tion. Our experiments show that acoustic discrete units not only achieve significantly faster convergence than mel-spectrograms but also produce speech with higher intelligibility. Addition- ally, we demonstrate that Cond...

    work page

  7. [7]

    F5-tts: A fairytaler that fakes fluent and faithful speech with flow matching,

    Y . Chen, Z. Niu, Z. Ma, K. Deng, C. Wang, J. Zhao, K. Yu, and X. Chen, “F5-tts: A fairytaler that fakes fluent and faithful speech with flow matching,” arXiv preprint arXiv:2410.06885, 2024

    arXiv 2024

  8. [9]

    Bigvgan: A universal neural vocoder with large-scale training,

    S.-g. Lee, W. Ping, B. Ginsburg, B. Catanzaro, and S. Yoon, “Bigvgan: A universal neural vocoder with large-scale training,” arXiv preprint arXiv:2206.04658, 2022

    arXiv 2022

Show all 34 references
  1. [10]

    Data augmentation using healthy speech for dysarthric speech recognition

    B. Vachhani, C. Bhat, and S. K. Kopparapu, “Data augmentation using healthy speech for dysarthric speech recognition.” in Inter- speech, 2018, pp. 471–475

    2018

  2. [11]

    Intelligibility of modifications to dysarthric speech,

    J.-P. Hosom, A. B. Kain, T. Mishra, J. P. Van Santen, M. Fried-Oken, and J. Staehely, “Intelligibility of modifications to dysarthric speech,” in 2003 IEEE International Conference on Acoustics, Speech, and Signal Processing, 2003. Proceed- ings.(ICASSP’03)., vol. 1. IEEE, 200...

    2003

  3. [12]

    Improving dysarthric speech intelligi- bility using cycle-consistent adversarial training,

    S. H. Yang and M. Chung, “Improving dysarthric speech intelligi- bility using cycle-consistent adversarial training,” arXiv preprint arXiv:2001.04260, 2020

    2001 arXiv

  4. [13]

    Phoneme-discriminative features for dysarthric speech conversion

    R. Aihara, T. Takiguchi, and Y . Ariki, “Phoneme-discriminative features for dysarthric speech conversion.” in Interspeech, 2017, pp. 3374–3378

    2017

  5. [14]

    Improving the efficiency of dysarthria voice conversion system based on data augmentation,

    W.-Z. Zheng, J.-Y . Han, C.-Y . Chen, Y .-J. Chang, and Y .-H. Lai, “Improving the efficiency of dysarthria voice conversion system based on data augmentation,” IEEE Transactions on Neural Sys- tems and Rehabilitation Engineering , vol. 31, pp. 4613–4623, 2023

    2023

  6. [15]

    Generative adver- sarial networks,

    I. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde- Farley, S. Ozair, A. Courville, and Y . Bengio, “Generative adver- sarial networks,”Communications of the ACM, vol. 63, no. 11, pp. 139–144, 2020

    2020

  7. [16]

    Generative adversarial networks: An overview,

    A. Creswell, T. White, V . Dumoulin, K. Arulkumaran, B. Sen- gupta, and A. A. Bharath, “Generative adversarial networks: An overview,” IEEE signal processing magazine , vol. 35, no. 1, pp. 53–65, 2018

    2018

  8. [17]

    Improving gans for speech enhance- ment,

    H. Phan, I. V . McLoughlin, L. Pham, O. Y . Ch ´en, P. Koch, M. De V os, and A. Mertins, “Improving gans for speech enhance- ment,” IEEE Signal Processing Letters , vol. 27, pp. 1700–1704, 2020

    2020

  9. [18]

    Cmgan: Conformer- based metric gan for speech enhancement,

    R. Cao, S. Abdulatif, and B. Yang, “Cmgan: Conformer- based metric gan for speech enhancement,” arXiv preprint arXiv:2203.15149, 2022

    2022 arXiv

  10. [19]

    Exploring speech en- hancement with generative adversarial networks for robust speech recognition,

    C. Donahue, B. Li, and R. Prabhavalkar, “Exploring speech en- hancement with generative adversarial networks for robust speech recognition,” in2018 IEEE international conference on acoustics, speech and signal processing (ICASSP). IEEE, 2018, pp. 5024– 5028

    2018

  11. [20]

    Segan: Speech enhancement generative adversarial network,

    S. Pascual, A. Bonafonte, and J. Serra, “Segan: Speech enhancement generative adversarial network,” arXiv preprint arXiv:1703.09452, 2017

    2017 arXiv

  12. [21]

    Unit- dsr: Dysarthric speech reconstruction system using speech unit normalization,

    Y . Wang, X. Wu, D. Wang, L. Meng, and H. Meng, “Unit- dsr: Dysarthric speech reconstruction system using speech unit normalization,” in ICASSP 2024-2024 IEEE International Con- ference on Acoustics, Speech and Signal Processing (ICASSP) . IEEE, 2024, pp. 12 306–12 310

    2024

  13. [22]

    Community-supported shared infrastructure in support of speech accessibility,

    M. Hasegawa-Johnson, X. Zheng, H. Kim, C. Mendes, M. Dickin- son, E. Hege, C. Zwilling, M. M. Channell, L. Mattie, H. Hodges et al., “Community-supported shared infrastructure in support of speech accessibility,” Journal of Speech, Language, and Hearing Research, vol. 67, no. ...

    2024

  14. [23]

    Self-supervised speech representations are more phonetic than semantic,

    K. Choi, A. Pasad, T. Nakamura, S. Fukayama, K. Livescu, and S. Watanabe, “Self-supervised speech representations are more phonetic than semantic,” arXiv preprint arXiv:2406.08619, 2024

    2024 arXiv

  15. [24]

    Hubert: Self-supervised speech represen- tation learning by masked prediction of hidden units,

    W.-N. Hsu, B. Bolte, Y .-H. H. Tsai, K. Lakhotia, R. Salakhutdi- nov, and A. Mohamed, “Hubert: Self-supervised speech represen- tation learning by masked prediction of hidden units,”IEEE/ACM transactions on audio, speech, and language processing, vol. 29, pp. 3451–3460, 2021

    2021

  16. [25]

    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. Xiaoet al., “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

    2022

  17. [26]

    Comparative layer-wise anal- ysis of self-supervised speech models,

    A. Pasad, B. Shi, and K. Livescu, “Comparative layer-wise anal- ysis of self-supervised speech models,” in ICASSP 2023-2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2023, pp. 1–5

    2023

  18. [27]

    A comparison of discrete and soft speech units for improved voice conversion,

    B. Van Niekerk, M.-A. Carbonneau, J. Za ¨ıdi, M. Baas, H. Seut ´e, and H. Kamper, “A comparison of discrete and soft speech units for improved voice conversion,” in ICASSP 2022-2022 IEEE In- ternational Conference on Acoustics, Speech and Signal Process- ing (ICASSP). IEEE, 20...

    2022

  19. [28]

    An information-theoretic analysis of self-supervised discrete repre- sentations of speech,

    B. M. Abdullah, M. M. Shaik, B. M ¨obius, and D. Klakow, “An information-theoretic analysis of self-supervised discrete repre- sentations of speech,” arXiv preprint arXiv:2306.02405, 2023

    2023 arXiv

  20. [29]

    Enhancing dysarthric speech recognition through sepformer and hierarchical attention network models with mul- tistage transfer learning,

    R. Vinotha, D. Hepsiba, L. Vijay Anand, J. Andrew, and R. Jen- nifer Eunice, “Enhancing dysarthric speech recognition through sepformer and hierarchical attention network models with mul- tistage transfer learning,” Scientific Reports , vol. 14, no. 1, p. 29455, 2024

    2024

  21. [30]

    Lib- rispeech: an asr corpus based on public domain audio books,

    V . Panayotov, G. Chen, D. Povey, and S. Khudanpur, “Lib- rispeech: an asr corpus based on public domain audio books,” in 2015 IEEE international conference on acoustics, speech and signal processing (ICASSP). IEEE, 2015, pp. 5206–5210

    2015

  22. [31]

    Gigaspeech: An evolving, multi-domain asr corpus with 10,000 hours of transcribed audio,

    G. Chen, S. Chai, G. Wang, J. Du, W.-Q. Zhang, C. Weng, D. Su, D. Povey, J. Trmal, J. Zhang et al., “Gigaspeech: An evolving, multi-domain asr corpus with 10,000 hours of transcribed audio,” arXiv preprint arXiv:2106.06909, 2021

    2021 arXiv

  23. [32]

    V oxpopuli: A large-scale multilingual speech corpus for representation learn- ing, semi-supervised learning and interpretation,

    C. Wang, M. Riviere, A. Lee, A. Wu, C. Talnikar, D. Haz- iza, M. Williamson, J. Pino, and E. Dupoux, “V oxpopuli: A large-scale multilingual speech corpus for representation learn- ing, semi-supervised learning and interpretation,” arXiv preprint arXiv:2101.00390, 2021

    2021 arXiv

  24. [33]

    Flow matching for generative modeling,

    Y . Lipman, R. T. Q. Chen, H. Ben-Hamu, M. Nickel, and M. Le, “Flow matching for generative modeling,” 2023. [Online]. Available: https://arxiv.org/abs/2210.02747

    2023 arXiv

  25. [34]

    Decoupled weight decay regularization,

    I. Loshchilov, “Decoupled weight decay regularization,” arXiv preprint arXiv:1711.05101, 2017

    2017 arXiv

  26. [35]

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

    A. Baevski, Y . Zhou, A. Mohamed, and M. Auli, “wav2vec 2.0: A framework for self-supervised learning of speech repre- sentations,” Advances in neural information processing systems , vol. 33, pp. 12 449–12 460, 2020

    2020

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.