Pith. sign in

REVIEW 3 major objections 5 minor 40 references

A Self-Training Approach for Whisper to Enhance Long Dysarthric Speech Recognition

T0 review · 3 major / 5 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 This paper claims that Whisper's recognition of long dysarthric speech improves from 6.84% to 5.55% WER by iteratively segmenting long audio, filtering segments against reference transcripts, and fine-tuning on the accepted segments.

desk verdict Solid challenge system with real gains, but the 'self-training' label does not survive contact with the paper: the method needs reference transcripts at both the filter and label steps. read the letter →

arxiv 2506.22810 v1 pith:ZZ3CWQZY submitted 2025-06-28 cs.SD eess.AS

Shiyao Wang , Jiaming Zhou , Shiwan Zhao , Yong Qin This is my paper
classification cs.SDeess.AS
keywords dysarthricspeechrecognitionWhisperself-traininglong-formsegmentationAccessibilityProjectworderrorratesemanticscore
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 claims that a self-training loop built around Whisper can substantially reduce word error rates on long dysarthric speech. The method, called ST-SLDS, cuts long recordings into short segments using the model's own predicted timestamps, keeps only segments whose combined transcription matches the reference transcript with zero word errors or at least with no insertions or deletions, and then fine-tunes Whisper on those segments. Iterating this process three times lowered WER on the SAP 0430 test set from 6.84% to 5.55% and on the newer 1130 test set from 8.84% to 7.47%. If this works, it turns otherwise unusable long audio into training data and simultaneously teaches the model to handle the partial segments that arise during inference. The resulting system ranked second in both WER and semantic score in the SAP Challenge.

What carries the argument

The load-bearing mechanism is the teacher model's predicted timestamps plus a reference-based heuristic filter. Long audio is segmented at the teacher's predicted word timestamps, each segment is transcribed, and the concatenated hypothesis is compared with the reference. Only utterances with WER=0 or with zero insertions and deletions (I=D=0) are accepted; the latter are re-tested in later iterations, and labels are produced either from the model's prediction (Algorithm a) or from the reference transcript aligned by word count (Algorithm b). This selection converts long dysarthric audio into short fine-tuning segments that simulate the incomplete inputs seen at inference, and the iterative teacher-student loop grows the training set while keeping the model aligned with its own segmentation behavior.

What would settle it

One could take long dysarthric recordings and withhold their references, then run the same teacher-student loop using only pseudo-label confidence (such as average log-probability) instead of WER equals zero or I equals D equals zero; if the WER improvement over baseline disappears or shrinks to the gain from merely adding unlabeled segments, the reference-based filter is the load-bearing component. Alternatively, shuffling or corrupting the reference words before filtering while keeping the same segmentation timestamps should collapse the reported gains if the labels are what matter, and leave them unchanged if segmentation alone drives the improvement.

Watch

Extended reading notes

Core claim

The central claim is that Whisper's performance on long dysarthric speech is limited both by encoder truncation of long spectrograms and by a mismatch between training on complete short utterances and inference on segmented partial ones, and that both can be addressed by a reference-filtered self-training scheme. Using fine-tuned Whisper large-v3 as teacher, the authors segment long dysarthric recordings at predicted timestamps, reconstruct the hypothesis by concatenating segment predictions, and compute WER against the reference. Segments are accepted when the reconstructed WER is zero or when insertion and deletion errors are both zero; accepted segments are labeled either with the model's own prediction (Algorithm a) or with reference words extracted to match segment length (Algorithm b). Repeating this for three iterations with the F-U text format, VAD-based segmentation, and beam size 10 at inference reduced 0430Test WER from 6.84% to 5.55% and 1130Test WER from 8.84% to 7.47%; a fourth iteration caused a small rebound attributed to duplicate data and overfitting. Adding the dev partition into training further improved 0430Test to 2.56% but degraded 1130Test, which the authors attribute to changed speaker distribution. On the challenge's inaccessible test sets the best submitted models achieved the second-best WER and SemScore.

Load-bearing premise

The pipeline needs the reference transcript of every long dysarthric recording to decide which segments to keep and how to label them, so the method is supervised in practice; without references, none of the filtering conditions can be computed and the self-training loop cannot start.

Editorial extensions

If this is right

  • Long dysarthric recordings with references can be recycled as training data without manual segmentation or transcription, increasing the pool by roughly 10 to 15 percent of samples per iteration.
  • Training on model-segmented partial utterances reduces the training-inference mismatch, so gains persist on held-out test sets such as 1130Test.
  • The benefit saturates around three iterations: a fourth iteration adds little data and slightly hurts performance due to duplicates and overfitting.
  • Choosing inference settings to match the teacher, including beam size 10, VAD-S segmentation, and the F-U text format, contributes nearly as much as the self-training itself.
  • The method is speaker- and text-independent in the sense of the SAP Challenge, since the added segments come from many speakers and topics.

Reading between the lines

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

  • This is better described as supervised segmentation-and-selection augmentation than self-training, because the filter and the Algorithm (b) labels both require the reference transcript; a true semi-supervised variant would need a proxy for WER such as prediction confidence, and could test whether reference-free filtering retains the gains.
  • The I=D=0 condition is an unusual choice: it admits substitution errors while excluding insertions and deletions, effectively selecting segments whose word boundaries and duration align with the reference; one could test whether this boundary alignment, rather than the label quality, drives the improvement.
  • The same two-stage filter could be applied to Whisper's long-speech processing in other domains, such as lecture or medical dictation, where transcripts are often available but segmentation is not.
  • Because the fourth iteration rebound is attributed to duplicate data, a deduplication step that removes segments already present from earlier iterations might extend the gains beyond three iterations.
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

3 major / 5 minor

Summary. The paper targets speaker-independent and text-independent dysarthric speech recognition on the Speech Accessibility Project (SAP) dataset. The authors compare several pretrained ASR architectures (HuBERT, ESPnet, Wenet, Whisper) fine-tuned on short SAP speech segments, find Whisper large-v3 with a VAD-based segmentation inference strategy to work best, and then propose an iterative method, ST-SLDS, that segments long dysarthric utterances into shorter training segments and adds them to the fine-tuning data. They report consistent WER reductions on an internal held-out set (1130Test) and on the SAP Challenge hidden Test 1 and Test 2 sets, culminating in a second-place finish in both WER and SemScore.

Significance. If the method is understood as a supervised segmentation-and-selection augmentation scheme, the empirical contribution is solid: the paper provides a useful comparison of modern ASR backbones on dysarthric speech, documents the benefit of segmentation-based inference for Whisper, and shows that adding segmented long utterances from the same labeled training partition improves performance on held-out and hidden sets. The use of the hidden Challenge tests for final evaluation is a strength, as is the clear reporting of training-set sizes and iteration-by-iteration results. However, the central claimed contribution -- a self-training approach that converts long dysarthric speech into usable training material without manual annotation -- is not supported by the described pipeline, because the filtering and labeling steps require the reference transcripts of the long audio. This mischaracterization affects the title, abstract, introduction, and method framing, and it needs to be corrected before the paper can be assessed on its actual merits.

major comments (3)
  1. [Section 2, Figures 1-2] The method is not self-training as described. The WER=0 and I=D=0 filters in Section 2 are computed by comparing the reconstructed hypothesis of the long utterance against its reference transcription, and Algorithm (b) in Figure 2 labels segments by extracting the matching number of words from that same reference. Because the long utterances are taken from the SAP0430 processed training partition, which already has reference transcripts (Section 3.1), the pipeline does not use unlabeled audio at all. It is a supervised segmentation-and-selection augmentation method. The paper should either rename the method and revise the abstract accordingly, or, if 'self-training' is retained, provide a version of the algorithm that does not require reference transcripts for filtering and labeling.
  2. [Table 3, rows 'Add dev'] The 'Add dev' rows add the SAP0430 processed dev partition, which includes 0430Test, into the training set. The paper itself states that 0430Test is then no longer a reference point, yet Table 3 still reports 0430Test WER and SemScore in the same columns as the genuinely held-out settings. Presenting these numbers in the same evaluation columns is misleading; they should be removed or clearly separated and labeled as training-set diagnostics rather than test results.
  3. [Section 3.3, Table 3] No statistical significance or variance is reported for the core ST-SLDS improvements. The differences between consecutive iterations are often small (for example, 5.5628 to 5.5533 on 0430Test between the 2nd and 3rd iterations), and the paper's claim that WER 'steadily decreased' over the first three iterations would be more convincing with confidence intervals, significance tests, or repeated fine-tuning runs. This is particularly important because the method's effectiveness is a central claim of the paper.
minor comments (5)
  1. [Figure 1] The figure contains the typo 'Hyo' for 'hypothesis', and the term 'Un-segmented long Dysarthric Speech A' is not defined precisely; please clarify whether A consists of the long utterances from the SAP0430 training partition and state how they were selected.
  2. [Section 3.2.1] The training duration limits are described ambiguously. The text says non-Whisper models are limited to 30 seconds, then says Whisper large-v3 is limited to 15 seconds, but it is not stated whether Lmax in Eq. (1) is set to 15 or 30 for the Whisper models used in the E-S experiments; please specify the value used for each model.
  3. [Section 3.2.2, Table 2] The checkmark-and-number format of Table 2 is hard to parse. Please add a sentence in the text or a footnote explaining exactly which dimensions each row corresponds to (for example, which rows use F-U, VAD-S, beam size 10, and prompt length) so the reader can re-derive the configuration of the final system.
  4. [Section 3.3] The sentence 'We submitted the models with relatively superior performance under both settings' is vague. Please specify the exact iteration and filtering condition of each submitted model and the reason for choosing it over the apparently better 3rd-iteration + I=D=0 + Add dev model on the hidden sets.
  5. [Section 1] The related-work discussion of self-training (references [25]-[31]) describes pseudo-labeling on unlabeled data, which highlights the discrepancy with the proposed method; the text should acknowledge this discrepancy explicitly and reposition the contribution as a segmentation-based data augmentation rather than semi-supervised learning.

Circularity Check

2 steps flagged · score 4.0 of 10

The claimed self-training loop is, by construction, a supervised segmentation-and-selection of reference-annotated long speech: WER=0 selection forces pseudo-labels to equal the reference, and I=D=0 labels are extracted from the reference. Hidden-test WER gains remain externally valid.

  1. self definitional [Section 2, Figures 1 and 2; Section 3.1]
    "Self-training, a common semi-supervised learning method [25, 26, 27, 28, 29, 30], is employed when labeled data is less and unlabeled speech data is abundant. ... (1) We reconstruct a prediction for the original long speech by combining the predictions of its segments, and then calculate its WER against the reference transcription."

    The premise of Section 2 is a self-training/semi-supervised loop on 'un-segmented long dysarthric speech A', yet the screening step (1) computes the long-utterance WER against the reference transcript, and Section 3.1 states the SAP0430 processed training partition is already reference-annotated ('Our models are trained using disfluency-removed transcripts'). Therefore the supposedly unlabeled long audio is labeled. A segment passes WER=0 only when the teacher hypothesis equals the reference, so Algorithm (a)'s 'corresponding prediction as the label' is, by construction, exactly the reference words for that segment.

  2. fitted input called prediction [Section 2, Figure 2, Algorithm (b)]
    "Data filtered by the “I = D = 0” condition uses Algorithm (b), also segmenting at the predicted timestamps but generating labels by extracting the same number of words from the reference transcription, matching the segment's position in the original speech."

    Algorithm (b) is used for I=D=0 segments. Its labels are not pseudo-labels: they are produced by 'extracting the same number of words from the reference transcription'. The training signal in this branch is literally the reference text, and the only model-derived part is the timestamp used for segmentation. Calling this 'annotated it using a self-training method' renames supervised reference labeling as self-training. The measured WER/SemScore gains on hidden tests are not invalidated, but the contribution's mechanism—pseudo-labeling of unlabeled long dysarthric speech—is not what the pipeline implements; it reduces to supervised segmentation plus reference-derived labels.

full rationale

The paper's external evaluation is not circular: 0430Test, 1130Test, and hidden Test 1/Test 2 are independent benchmarks, and the measured WER gains are real evidence. The circularity concern is confined to the claimed self-training mechanism. Two concrete reductions are present: (1) WER=0 selection guarantees the teacher hypothesis equals the reference, so the pseudo-labels used in Algorithm (a) are identical to the reference by construction; (2) Algorithm (b) labels segments by extracting words from the reference transcript. In both cases the 'unlabeled long dysarthric speech A' is in fact reference-annotated (Section 3.1 states the training partition and the model's use of disfluency-removed transcripts). The pipeline therefore requires reference transcripts and cannot be applied to genuinely unlabeled speech as claimed in the self-training framing. This is more than a wording issue because the title, abstract, and introduction present the method as a self-training/semi-supervised contribution; however, the hidden-test performance is independent of that framing. Score 4 reflects partial circularity in the label-construction claim, not in the benchmark results. No load-bearing self-citation is present.

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

The method introduces no new model components or entities, but it does rely on several domain assumptions: that reference transcripts are always available for 'self-training', that Whisper timestamps are accurate enough for segmentation, and that a 15-second training limit transfers to longer inference segments. The free parameters are standard hyperparameters and a stopping rule, plus the heuristic filtering thresholds.

free parameters (4)
  • Whisper max training duration Lmax = 15 seconds
    Chosen from the SAP0430 duration distribution as twice the peak concentration to balance GPU utilization and training length; affects what gets segmented.
  • Beam size = 10
    Selected empirically on 0430Test as optimal among 5, 10, 20, 30.
  • Number of ST-SLDS iterations = 3
    Stopped after the third iteration because the fourth added minimal data and performance rebounded; chosen post hoc based on validation.
  • Filtering conditions WER=0 and I=D=0
    Heuristic thresholds for accepting pseudo-labels; not derived from theory.
assumptions (3)
  • domain assumption Reference transcripts are available for the long speech used in self-training
    Section 2 and Figure 2 rely on computing WER against the reference of long audio to filter data; Algorithm (b) labels segments from the reference. Without this, the method cannot run.
  • domain assumption Whisper's predicted timestamps align well enough with speech boundaries
    Algorithms (a) and (b) segment at teacher-model timestamps; inaccurate timestamps would create misaligned audio-label pairs. The paper does not validate timestamp quality on dysarthric speech.
  • domain assumption Fine-tuning on up to 15-second segments transfers to inference segments near 30 seconds
    Section 3.2.1 limits training to 15 seconds while inference VAD-S can create segments approaching 30 seconds, implicitly assuming no harmful train-inference gap.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Self-Training Approach for Whisper to Enhance Long Dysarthric Speech Recognition." pith.science (2026). https://pith.science/paper/ZZ3CWQZY

@misc{pith2026250622810,
  author       = {Pith},
  title        = {Pith review of: A Self-Training Approach for Whisper to Enhance Long Dysarthric Speech Recognition},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZZ3CWQZY}},
  note         = {Machine review of arXiv:2506.22810}
}
read the original abstract

Dysarthric speech recognition (DSR) enhances the accessibility of smart devices for dysarthric speakers with limited mobility. Previously, DSR research was constrained by the fact that existing datasets typically consisted of isolated words, command phrases, and a limited number of sentences spoken by a few individuals. This constrained research to command-interaction systems and speaker adaptation. The Speech Accessibility Project (SAP) changed this by releasing a large and diverse English dysarthric dataset, leading to the SAP Challenge to build speaker- and text-independent DSR systems. We enhanced the Whisper model's performance on long dysarthric speech via a novel self-training method. This method increased training data and adapted the model to handle potentially incomplete speech segments encountered during inference. Our system achieved second place in both Word Error Rate and Semantic Score in the SAP Challenge.

Figures

Figures reproduced from arXiv: 2506.22810 by the authors.

Figure 1
Figure 1. The overview of our self-training approach to segment long dysarthric speech (ST-SLDS). ‘Hyo’: ‘hypothesis’. Algo (a) and Algo (b) see [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Segmentation algorithm (a) and (b). ‘wn’ represents the number of words. ‘Ref ’: ‘Reference’. ‘Algo’: ‘Algorithm’. complete sentences even from partial inputs. Liu et al. [24] observed similar mismatches in speech recognition. Both stud￾ies addressed this by incorporating “partial sequence” data into training. Following their approach, we attempted to incorporate segmented data to mitigate the training-inference inc… view at source ↗
Figure 3
Figure 3. The duration distribution of the training data in SAP0430 processed. Training was conducted under the Fairseq framework3 , based on the hyperparameters outlined in the base 10h.yaml configu￾ration. We set the learning rate to 10−5 and the warmup steps to 32,000, halting the training when the validation set loss did not decrease over 10 consecutive epochs. ESPnet Pretrain: We fine-tuned two separate models within the… view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

40 extracted references · 33 canonical work pages

  1. [1]

    A Self-Training Approach for Whisper to Enhance Long Dysarthric Speech Recognition

    Introduction Dysarthria, a speech disorder stemming from neurological con- ditions like Parkinson’s disease and cerebral palsy, affects vo- cal control and leads to speech impediments such as stutter- ing, abnormal pauses, altered prosody, and mispronunciation. These characteristics pose significant challenges for automatic speech recognition (ASR) system...

    work page Pith review arXiv 2024

  2. [2]

    WER = 0” condition uses Algorithm (a), segmenting the speech at the predicted timestamps and using the corresponding prediction as the label. Data filtered by the “I = D = 0

    Whisper self-training optimization We propose a self-training approach to segment long dysarthric speech (ST-SLDS, see Figure 1). First, we fine-tune the Whis- per model on short dysarthric speech data to create a teacher model. In subsequent iterations, the fine-tuning dataset includes segmented data from previous iterations. The teacher model then perfo...

    work page

  3. [3]

    I = D = 0

    Experiments 3.1. Dataset The SAP Challenge data comprises the following components: (1) SAP0430 processed: This dataset has been preprocessed through 16kHz audio resampling and text regularization and is divided into training (290.4 hours, 131,420 samples, 369 speak- ers) and development (dev) (43.6 hours, 19,275 samples, 55 speakers) partitions. The dev ...

    work page 2024

  4. [4]

    By segmenting the data, we sim- ulate the incomplete speech segments encountered during in- ference

    Conclusions To improve Whisper’s recognition of long dysarthric speech, we propose a self-training method that segments and labels long speech, enabling its use for fine-tuning and mitigating training-inference mismatch. By segmenting the data, we sim- ulate the incomplete speech segments encountered during in- ference. Ultimately, our constructed speaker...

    work page

  5. [5]

    Acknowledgements This work has been supported in part by NSF China (Grant No.62271270)

    work page

  6. [6]

    The torgo database of acoustic and articulatory speech from speakers with dysarthria,

    F. Rudzicz, A. K. Namasivayam, and T. Wolff, “The torgo database of acoustic and articulatory speech from speakers with dysarthria,” Language resources and evaluation, vol. 46, 2012

    work page 2012

  7. [7]

    On the impact of dysarthric speech on contemporary asr cloud platforms,

    L. De Russis and F. Corno, “On the impact of dysarthric speech on contemporary asr cloud platforms,”Journal of Reliable Intelligent Environments, vol. 5, pp. 163–172, 2019

    work page 2019

  8. [8]

    Robust speech recognition via large-scale weak su- pervision,

    A. Radford, J. W. Kim, T. Xu, G. Brockman, C. McLeavey, and I. Sutskever, “Robust speech recognition via large-scale weak su- pervision,” in ICML. PMLR, 2023, pp. 28 492–28 518

    work page 2023

Show all 40 references
  1. [9]

    Google usm: Scaling auto- matic speech recognition beyond 100 languages,

    Y . Zhang, W. Han, J. Qin, Y . Wang, A. Bapna, Z. Chen, N. Chen, B. Li, V . Axelrod, G. Wang et al. , “Google usm: Scaling auto- matic speech recognition beyond 100 languages,” arXiv preprint arXiv:2303.01037, 2023

    2023 arXiv

  2. [10]

    Dysarthric speech database for universal access research,

    H. Kim, M. Hasegawa-Johnson, A. Perlman, J. Gunderson, T. S. Huang, K. Watkin, and S. Frame, “Dysarthric speech database for universal access research,” in Ninth Annual Conference of the International Speech Communication Association , 2008

    2008

  3. [11]

    Easycall corpus: a dysarthric speech dataset,

    R. Turrisi, A. Braccia, M. Emanuele, S. Giulietti, M. Pugliatti, M. Sensi, L. Fadiga, and L. Badino, “Easycall corpus: a dysarthric speech dataset,” arXiv preprint arXiv:2104.02542, 2021

    2021 arXiv

  4. [12]

    Pb-lrdwws system for the slt 2024 low-resource dysarthria wake-up word spotting chal- lenge,

    S. Wang, J. Zhou, S. Zhao, and Y . Qin, “Pb-lrdwws system for the slt 2024 low-resource dysarthria wake-up word spotting chal- lenge,” in 2024 IEEE Spoken Language Technology Workshop (SLT). IEEE, 2024, pp. 586–591

    2024

  5. [13]

    Synthesizing dysarthric speech using multi-speaker tts for dysarthric speech recognition,

    M. Soleymanpour, M. T. Johnson, R. Soleymanpour, and J. Berry, “Synthesizing dysarthric speech using multi-speaker tts for dysarthric speech recognition,” in ICASSP. IEEE, 2022

    2022

  6. [14]

    Training data augmentation for dysarthric automatic speech recognition by text- to-dysarthric-speech synthesis,

    W.-Z. Leung, M. Cross, A. Ragni, and S. Goetze, “Training data augmentation for dysarthric automatic speech recognition by text- to-dysarthric-speech synthesis,” in Interspeech, 2024

    2024

  7. [15]

    Per- sonalizing asr for dysarthric and accented speech with limited data,

    J. Shor, D. Emanuel, O. Lang, O. Tuval, M. Brenner, J. Cattiau, F. Vieira, M. McNally, T. Charbonneau, M. Nollstadtet al., “Per- sonalizing asr for dysarthric and accented speech with limited data,” arXiv preprint arXiv:1907.13511, 2019

    1907 arXiv

  8. [16]

    A survey of technologies for au- tomatic dysarthric speech recognition,

    Z. Qian, K. Xiao, and C. Yu, “A survey of technologies for au- tomatic dysarthric speech recognition,” EURASIP Journal on Au- dio, Speech, and Music Processing , vol. 2023, no. 1, p. 48, 2023

    2023

  9. [17]

    Enhancing dysarthric speech recognition for unseen speakers via prototype- based adaptation,

    S. Wang, S. Zhao, J. Zhou, A. Kong, and Y . Qin, “Enhancing dysarthric speech recognition for unseen speakers via prototype- based adaptation,” in Interspeech 2024, 2024, pp. 1305–1309

    2024

  10. [18]

    Espnet: End-to-end speech processing toolkit,

    S. Watanabe, T. Hori, S. Karita, T. Hayashi, J. Nishitoba, Y . Unno, N. E. Y . Soplin, J. Heymann, M. Wiesner, N. Chen et al. , “Espnet: End-to-end speech processing toolkit,” arXiv preprint arXiv:1804.00015, 2018

    2018 arXiv

  11. [19]

    Community-supported shared infrastructure in support of speech accessibility,

    M. Hasegawa-Johnson, X. Zheng, H. Kim, C. Mendes, M. Dickinson, E. Hege, C. Zwilling, M. M. Channell, L. Mattie, H. Hodges, L. Ramig, M. Bellard, M. Shebanek, L. Sar, K. Kalgaonkar, D. Frerichs, J. P. Bigham, L. Findlater, C. Lea, S. Herrlinger, P. Korn, S. Abou-Zahra, R. Heyw...

  12. [20]

    M2r-whisper: Multi-stage and multi-scale retrieval augmentation for enhancing whisper,

    J. Zhou, S. Zhao, J. He, H. Wang, W. Zeng, Y . Chen, H. Sun, A. Kong, and Y . Qin, “M2r-whisper: Multi-stage and multi-scale retrieval augmentation for enhancing whisper,” arXiv preprint arXiv:2409.11889, 2024

    2024 arXiv

  13. [21]

    Fine-tuning automatic speech recognition for people with parkinson’s: An ef- fective strategy for enhancing speech technology accessibility,

    X. Zheng, B. Phukon, and M. Hasegawa-Johnson, “Fine-tuning automatic speech recognition for people with parkinson’s: An ef- fective strategy for enhancing speech technology accessibility,” arXiv preprint arXiv:2409.19818, 2024

    2024 arXiv

  14. [22]

    Robust cross-etiology and speaker-independent dysarthric speech recognition,

    S. Singh, Q. Wang, Z. Zhong, C. Mendes, M. Hasegawa- Johnson, W. Abdulla, and S. R. Shahamiri, “Robust cross-etiology and speaker-independent dysarthric speech recognition,” arXiv preprint arXiv:2501.14994, 2025

    2025 arXiv

  15. [23]

    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

  16. [24]

    Hubert: Self-supervised speech rep- resentation learning by masked prediction of hidden units,

    W.-N. Hsu, B. Bolte, Y .-H. H. Tsai, K. Lakhotia, R. Salakhut- dinov, and A. Mohamed, “Hubert: Self-supervised speech rep- resentation learning by masked prediction of hidden units,” IEEE/ACM Transactions on Audio, Speech, and Language Pro- cessing, vol. 29, pp. 3451–3460, 2021

    2021

  17. [25]

    Self-training for end-to-end speech recognition,

    J. Kahn, A. Lee, and A. Hannun, “Self-training for end-to-end speech recognition,” in ICASSP. IEEE, 2020, pp. 7084–7088

    2020

  18. [26]

    Wenet: Production oriented stream- ing and non-streaming end-to-end speech recognition toolkit,

    Z. Yao, D. Wu, X. Wang, B. Zhang, F. Yu, C. Yang, Z. Peng, X. Chen, L. Xie, and X. Lei, “Wenet: Production oriented stream- ing and non-streaming end-to-end speech recognition toolkit,” arXiv preprint arXiv:2102.01547, 2021

    2021 arXiv

  19. [27]

    Lessons from building acoustic models with a million hours of speech,

    S. H. K. Parthasarathi and N. Strom, “Lessons from building acoustic models with a million hours of speech,” in ICASSP. IEEE, 2019, pp. 6670–6674

    2019

  20. [28]

    Whisperx: Time- accurate speech transcription of long-form audio,

    M. Bain, J. Huh, T. Han, and A. Zisserman, “Whisperx: Time- accurate speech transcription of long-form audio,” arXiv preprint arXiv:2303.00747, 2023

    2023 arXiv

  21. [29]

    Universal speech disorder recognition: Towards a foun- dation model for cross-pathology generalisation,

    G. Sanguedolce, D.-C. Gruia, S. Brook, P. Naylor, and F. Geran- mayeh, “Universal speech disorder recognition: Towards a foun- dation model for cross-pathology generalisation,” in Advance- ments In Medical F oundation Models: Explainability, Robustness, Security, and Beyond

  22. [30]

    Low-latency neural speech translation,

    J. Niehues, N.-Q. Pham, T.-L. Ha, M. Sperber, and A. Waibel, “Low-latency neural speech translation,” arXiv preprint arXiv:1808.00491, 2018

    2018 arXiv

  23. [31]

    The filtered data is then combined with the labeled data to train a student model, which subsequently becomes the new teacher, and the process is repeated iteratively

    (e.g., frame-level, utterance-level) or through heuristics [25]. The filtered data is then combined with the labeled data to train a student model, which subsequently becomes the new teacher, and the process is repeated iteratively. In the context of improving Whisper’s perfor...

  24. [32]

    Low-latency sequence-to- sequence speech recognition and translation by partial hypothesis selection,

    D. Liu, G. Spanakis, and J. Niehues, “Low-latency sequence-to- sequence speech recognition and translation by partial hypothesis selection,” arXiv preprint arXiv:2005.11185, 2020

    2005 arXiv

  25. [33]

    Large-scale semi- supervised training in deep learning acoustic model for asr,

    Y . Long, Y . Li, S. Wei, Q. Zhang, and C. Yang, “Large-scale semi- supervised training in deep learning acoustic model for asr,”IEEE Access, vol. 7, pp. 133 615–133 627, 2019

    2019

  26. [35]

    Making more of little data: Improving low-resource automatic speech recognition using data augmentation,

    M. Bartelds, N. San, B. McDonnell, D. Jurafsky, and M. Wieling, “Making more of little data: Improving low-resource automatic speech recognition using data augmentation,” arXiv preprint arXiv:2305.10951, 2023

    2023 arXiv

  27. [36]

    End-to- end asr: from supervised to semi-supervised learning with modern architectures,

    G. Synnaeve, Q. Xu, J. Kahn, T. Likhomanenko, E. Grave, V . Pratap, A. Sriram, V . Liptchinsky, and R. Collobert, “End-to- end asr: from supervised to semi-supervised learning with modern architectures,” arXiv preprint arXiv:1911.08460, 2019

    1911 arXiv

  28. [37]

    Madi: Inter-domain matching and intra-domain discrimination for cross- domain speech recognition,

    J. Zhou, S. Zhao, N. Jiang, G. Zhao, and Y . Qin, “Madi: Inter-domain matching and intra-domain discrimination for cross- domain speech recognition,” in ICASSP 2023-2023 IEEE Inter- national Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2023, pp. 1–5

    2023

  29. [38]

    Semi-supervised training of deep neural networks,

    K. Vesel `y, M. Hannemann, and L. Burget, “Semi-supervised training of deep neural networks,” in2013 IEEE Workshop on Au- tomatic Speech Recognition and Understanding . IEEE, 2013, pp. 267–272

    2013

  30. [39]

    Reproducing whisper-style training using an open-source toolkit and publicly available data,

    Y . Peng, J. Tian, B. Yan, D. Berrebbi, X. Chang, X. Li, J. Shi, S. Arora, W. Chen, R. Sharma et al., “Reproducing whisper-style training using an open-source toolkit and publicly available data,” in 2023 IEEE Automatic Speech Recognition and Understanding Workshop (ASRU). IEE...

    2023

  31. [40]

    Assessing asr model quality on disordered speech using bertscore,

    J. Tobin, Q. Li, S. Venugopalan, K. Seaver, R. Cave, and K. Tomanek, “Assessing asr model quality on disordered speech using bertscore,” arXiv preprint arXiv:2209.10591, 2022

    2022 arXiv

  32. [2024]

    Available: https://pubs.asha.org/doi/abs/10.1044/ 2024 JSLHR-24-00122

    [Online]. Available: https://pubs.asha.org/doi/abs/10.1044/ 2024 JSLHR-24-00122

    2024

Pith tools

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