Pith. sign in

REVIEW 2 major objections 6 minor 48 references

Transliterated Zero-Shot Domain Adaptation for Automatic Speech Recognition

T0 review · 2 major / 6 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read The paper claims that transliterated zero-shot domain adaptation—using transliterations as pre-training labels so pre-training and fine-tuning labels share one writing system—preserves pre-trained domain knowledge and reduces…

desk verdict Useful zero-shot ASR adaptation method, but the headline comparison is confounded by extra supervised pre-training on the target language, so the label-consistency mechanism is not established. read the letter →

arxiv 2412.11185 v1 pith:NSAXR4Y2 submitted 2024-12-15 eess.AS cs.CLcs.SD

Han Zhu , Gaofeng Cheng , Qingwei Zhao , Pengyuan Zhang This is my paper · ORCID
classification eess.AScs.CLcs.SD
keywords automaticspeechrecognitionzero-shotdomainadaptationcross-lingualpre-trainingtransliterationpseudo-labelingknowledgepreservationcurriculumlearningwav2vec2.0
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

Zero-shot domain adaptation asks whether an ASR model can be moved to a new acoustic domain when the only available target-domain audio is in another language. This paper proposes transliterated ZSDA: decode the source-language target-domain audio with a target-language seed ASR model, use the resulting transliterations as pre-training labels, then fine-tune on the target language. Because transliterations and transcriptions are written with the same target-language tokens, the pre-trained representations change less during fine-tuning, so the domain knowledge learned from the source language survives. On Cantonese and Czech targets, the method reduces WER by 9.2% relative to a wav2vec 2.0 baseline, beats self-supervised ZSDA, and matches a supervised variant that uses ground-truth source-language transcripts.

What carries the argument

The central object is transliterated cross-lingual pre-training (transliterated XLPT). A transliteration is the source-language utterance rendered in target-language characters by greedy-decoding the output of a CTC-based target-language ASR model; by construction, pre-training labels and fine-tuning labels are drawn from the same token set. The procedure couples this with curriculum XLPT (self-supervised pre-training on both languages before transliteration-based pre-training), continuous pseudo-labeling (the teacher label set is refreshed as an EMA of the student), and a shared-hidden-layer architecture with separate classifiers per language to absorb the distribution gap between transcriptions and transliterations. The proposed BT-CTC metric measures label quality by synthesizing speech from a transliteration with target-language TTS and comparing it to the original source speech through a source-language ASR model.

What would settle it

Compare the generated transliterations, on a held-out set of source-language target-domain utterances with ground-truth transcripts, against a reference obtained by mapping those transcripts into the target writing system with a pronunciation dictionary; if the transliterations are near chance while the 9.2% relative WER gain still appears, the label-accuracy explanation is wrong and label-system consistency alone is doing the work.

Watch

Extended reading notes

Core claim

The central claim is that cross-lingual knowledge transfer succeeds to the extent that pre-training and fine-tuning labels are in the same writing system. Transliterations are generated by greedy-decoding source-language speech with a teacher model that is the exponential moving average of the student, so they are pseudo-labels in the target language's script; the student is trained with CTC loss on target transcriptions and source transliterations jointly. The paper's evidence that this works: representation-similarity (CCA) between pre-trained and fine-tuned models is higher for transliterated than self-supervised pre-training, the substitution-error component of WER drops, and the full method reaches an average WER of 40.2 versus 41.4 for self-supervised ZSDA and 44.4 for the wav2vec 2.0 baseline. The paper concludes that transliterated ZSDA outperforms self-supervised ZSDA and performs on par with supervised ZSDA without requiring source-language annotation.

Load-bearing premise

The method stands on the assumption that transliterations produced by a target-language seed ASR model decoding source-language audio are accurate enough to serve as training labels, a quality the authors themselves note cannot be guaranteed.

Editorial extensions

If this is right

  • Adaptation becomes possible without any target-domain audio in the target language or any transcriptions in the source language; unlabeled related-language audio plus a seed model is sufficient.
  • The 9.2% relative WER gain over the wav2vec 2.0 baseline and the consistent lead over self-supervised ZSDA indicate that transliteration-based labels transfer more domain knowledge than generic self-supervised pre-training.
  • Matching supervised ZSDA implies that ground-truth source-language transcriptions are not necessary to reach the same adaptation quality, which lowers the cost of domain adaptation in low-resource settings.
  • Source-language choice matters: close languages in the same family and in-domain source data are required; replacing either with a distant language or an out-of-domain source degrades WER.
  • Both curriculum pre-training and continuous pseudo-labeling improve transliteration quality, and better transliterations translate into lower WER, so label quality is a controllable driver of adaptation success.

Reading between the lines

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

  • A direct test this paper leaves open is whether the writing-system match itself, rather than label accuracy, carries part of the benefit: swapping in source-script pseudo-labels of equal phonetic accuracy would separate the two explanations.
  • The BT-CTC metric could be used as a cheap quality gate before full pre-training: computing it on a few hours of source-language target-domain audio would indicate whether transliterated ZSDA is likely to help.
  • The recipe should carry over to dialect or accent adaptation, where target-language audio is scarce but a related variety is abundant, provided the seed model can decode the source variety into the target variety's orthography.
  • A failure mode the paper does not explore is non-linguistic or heavily code-switched source-domain audio, where pronunciation-only transliterations lose content and would degrade the pseudo-labels.
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

2 major / 6 minor

Summary. The paper proposes transliterated zero-shot domain adaptation (ZSDA) for automatic speech recognition, a method that uses transliterations—source-language speech decoded into the target-language writing system—as pre-training labels for cross-lingual pre-training (XLPT), followed by target-language fine-tuning. The core claim is that using transliterations makes the pre-training and fine-tuning labels share one writing system, reducing representation change during fine-tuning and thereby preserving pre-trained domain knowledge. The method is evaluated on Cantonese (source: Mandarin) and Czech (source: Russian), reporting a 9.2% relative WER reduction over a wav2vec 2.0 baseline and consistent improvements over self-supervised ZSDA. The paper also introduces a curriculum XLPT scheme, continuous pseudo-labeling, a back-transliteration CTC (BT-CTC) quality metric, and several ablations.

Significance. The paper addresses a practically important problem: adapting ASR to a new domain when target-domain audio is available only in another language. The proposed method is novel in using transliteration-style labels as a bridge between languages and in avoiding manual annotation of the source language. The experimental design has notable strengths: two language pairs, public datasets, matched total update budgets, and ablations for each design choice (curriculum XLPT, continuous pseudo-labeling, separated classifiers, graphemic vs. phonemic supervision). The BT-CTC metric is a creative way to monitor pseudo-label quality. If the central mechanism were supported, the contribution would be significant. However, as detailed below, the main experimental comparison does not isolate the label-consistency mechanism, so the paper's central causal claim is currently not fully supported.

major comments (2)
  1. [Section 6.1 / Table 3 / Algorithm 1] The comparison against self-supervised ZSDA is confounded by the amount of supervised training on the labeled set L. In transliterated ZSDA, L is used with CTC supervision in the seeding stage (Algorithm 1 lines 10-12, Eq. (1)) and again in the pseudo-labeling stage (lines 15-16, Eq. (3)), for 30k additional updates before fine-tuning; self-supervised ZSDA uses L only in the 40k fine-tuning stage. Section 5.2 states that all methods have the same total pre-training updates, but it does not match the amount or type of supervision. The 1.2-point average WER advantage (40.2 vs 41.4) could therefore be due to the extra supervised exposure to L rather than to the consistency of transliteration labels. A control is needed: for example, self-supervised XLPT for 40k, followed by the same 10k supervised seeding on L and the same 20k combined supervised-L plus self-supervised-U pseudo-labeling schedule, then fine-tuning. Without such a control, the paper's central claim that label consistency preserves pre-trained knowledge is not established.
  2. [Section 6.2 / Figure 4] The CCA-similarity analysis is not independent of the confound above. The 'transliterated XLPT' model used in the comparison already includes 30k supervised CTC updates on L, so its representations are expected to be closer to the fine-tuned model even if transliteration labels were irrelevant. To support the claim that less representation change is due to label consistency, the comparison should hold the amount of supervised L updates fixed between the two pre-training methods (e.g., the control described above). As it stands, Figure 4 is consistent with the alternative explanation that additional supervised updates on L reduce the distance to the fine-tuned model.
minor comments (6)
  1. [Section 4.1 / Eqs. (1) and (3)] Equations (1) and (3) are missing closing parentheses for the set condition; for example, 'Lsup = CTC(y, fθt(a(x)), (x, y)∈ L' should be 'Lsup = CTC(y, fθt(a(x))), (x, y)∈ L'.
  2. [Table 3 header] The grouping 'ZSDA with labeled target domain data' for the Supervised ZSDA row is ambiguous, because that row uses labeled source-language data in the target domain rather than labeled target-language data; consider rewording to 'labeled source-language target-domain data'.
  3. [Figure 6] In the manuscript text provided, Figure 6 renders as '/uni0000...' sequences, making the token distribution plots unreadable; please ensure the figure uses proper character rendering.
  4. [Section 7 / Conclusion] The conclusion states that transliterated ZSDA 'has the same average performance with the supervised ZSDA', but Table 3 shows 40.2 vs 42.0, which is a small improvement rather than equality; consider saying it is comparable or slightly better.
  5. [Section 4.3 / BT-CTC metric] The BT-CTC metric is an indirect proxy for transliteration quality that depends on a source-language ASR model and a target-language TTS system; an external validation, such as human evaluation or comparison with a conventional transliteration system, would strengthen the claim that transliterations are accurate enough for the method to work, though the ablations in Tables 5-6 partially address this concern.
  6. [Section 6.3] The text 'both curriculum XLST and continuous pseudo-labeling' contains a typo: 'XLST' should be 'XLPT'.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the 9.2% WER gain is a measured empirical result, not constructed from the method's definitions; the label-consistency mechanism is separately supported by CCA similarity, t-SNE, ablations, and Appendix A.

full rationale

The central result is an empirical WER comparison (Table 3) against baselines with matched total update counts; no parameter is fitted to produce the reported 9.2% reduction. Transliteration generation (Eq. 4) uses a target-language teacher, and the BT-CTC quality metric (Eq. 7) uses an external source-language ASR; the metric is a proxy, not the training objective, so it does not make the central claim true by construction. The CCA similarity and t-SNE analyses in Section 6.2 provide independent evidence for the label-consistency mechanism, and Appendix A is a standard gradient-based argument. Self-citations (Zhu et al. 2022, 2023) support prior claims but are not load-bearing: the forgetting claim is also anchored to external work (Pasad et al. 2021) and to the paper's own measurements. The legitimate caveat is an experimental-design confound, not circularity: transliterated ZSDA gives L 30k supervised XLPT updates (10k seed + 20k pseudo-labeling) plus 40k fine-tuning, whereas self-supervised ZSDA gives L only the 40k supervised fine-tuning updates, so the comparison with self-supervised ZSDA does not fully isolate label consistency. This is a validity concern about attribution, not an equivalence between inputs and outputs, and does not affect the circularity score.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The paper's central claim rests on three empirical assumptions: that shared representations transfer across the chosen language pair, that transliterations produced by a target-language seed model are informative enough to pre-train on, and that label-consistency (same writing system) preserves pre-trained domain knowledge. None of these is derived from first principles; they are supported by ablations and prior work. There is no fitted parameter that manufactures the main result, but several hyperparameters (curriculum split, EMA decay alpha, learning rate) are chosen by hand and not all are reported.

free parameters (3)
  • Curriculum schedule split = 10k self-supervised + 30k transliterated updates
    The split between self-supervised curriculum pre-training and transliterated pre-training is chosen by hand (Section 5.2). The ablation in Table 6 shows removing the curriculum hurts WER by about 1.4 points on average, so the central result depends on this choice.
  • EMA decay factor alpha
    The teacher model in continuous pseudo-labeling is updated as an exponential moving average with decay alpha (Eq. 5), but the paper never states the alpha value, so the transliteration quality and hence the final WER depend on an unreported hyperparameter.
  • Learning rate and batch size = 3e-5, 25.6m audio samples
    These optimizer settings (Section 5.2) are standard for wav2vec 2.0 fine-tuning, but they are chosen without a sensitivity analysis; the reported gains are not shown to be robust to these choices.
assumptions (5)
  • domain assumption Representation extraction layers can be shared across languages while classifiers must be separate
    Invoked in Section 4.1 and Section 6.5 (shared-hidden-layer architecture from Huang et al. 2013). If representations are not actually shareable across the chosen language pair, the whole XLPT transfer fails.
  • domain assumption A close source language provides positive transfer and a distant language provides negative transfer
    Assumed in Section 3.2 and tested in Section 6.4 (Table 7), where English hurts and AISHELL-1 hurts. The method requires this to hold for the chosen language pair.
  • domain assumption The seed target-language ASR model, trained only on source-domain data, can decode source-language audio into meaningful graphemic transliterations
    This is the enabler of Eq. 4 in Section 4.1. The paper mitigates noise with curriculum and continuous pseudo-labeling but has no proof that the transliterations are accurate; BT-CTC only approximates quality.
  • standard math CTC loss is a sufficient training objective for cross-lingual pre-training and fine-tuning
    The paper uses CTC throughout (Eq. 1-3, 7), following standard ASR practice. This is a domain assumption from prior work, not introduced ad hoc.
  • domain assumption Pre-training on self-supervised labels first improves out-of-domain robustness
    Used in the curriculum XLPT (Section 4.2), relying on prior results such as Hsu et al. 2021a. This is an empirical assumption, and the paper's own ablation (Table 9) confirms it matters.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Transliterated Zero-Shot Domain Adaptation for Automatic Speech Recognition." pith.science (2026). https://pith.science/paper/NSAXR4Y2

@misc{pith2026241211185,
  author       = {Pith},
  title        = {Pith review of: Transliterated Zero-Shot Domain Adaptation for Automatic Speech Recognition},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NSAXR4Y2}},
  note         = {Machine review of arXiv:2412.11185}
}
read the original abstract

The performance of automatic speech recognition models often degenerates on domains not covered by the training data. Domain adaptation can address this issue, assuming the availability of the target domain data in the target language. However, such assumption does not stand in many real-world applications. To make domain adaptation more applicable, we address the problem of zero-shot domain adaptation (ZSDA), where target domain data is unavailable in the target language. Instead, we transfer the target domain knowledge from another source language where the target domain data is more accessible. To do that, we first perform cross-lingual pre-training (XLPT) to share domain knowledge across languages, then use target language fine-tuning to build the final model. One challenge in this practice is that the pre-trained knowledge can be forgotten during fine-tuning, resulting in sub-optimal adaptation performance. To address this issue, we propose transliterated ZSDA to achieve consistent pre-training and fine-tuning labels, leading to maximum preservation of the pre-trained knowledge. Experimental results show that transliterated ZSDA relatively decreases the word error rate by 9.2% compared with a wav2vec 2.0 baseline. Moreover, transliterated ZSDA consistently outperforms self-supervised ZSDA and performs on par with supervised ZSDA, proving the superiority of transliteration-based pre-training labels.

Figures

Figures reproduced from arXiv: 2412.11185 by the authors.

Figure 1
Figure 1. The diagram of the proposed ZSDA framework: cross-lingual pre-training (XLPT) and target language fine-tuning. The proposed [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Illustration transliterated XLPT. In this example, the target language is Cantonese, whereas the source language is Mandarin. To illustrate [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Computation procedure of BT-CTC loss. BT-CTC loss can measure the similarity between transliteration and source language speech. [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Illustration of representation similarity between the pre-trained and fine-tuned models with transliterated or self-supervised XLPT meth [PITH_FULL_IMAGE:figures/full_fig_p010_4.png]
Figure 5
Figure 5. Figure 5: Comparison of representation before and after fine-tuning for self-supervised and transliterated XLPT models. The same color denotes [PITH_FULL_IMAGE:figures/full_fig_p011_5.png]
Figure 6
Figure 6. Figure 6: Comparison of token distributions in transcriptions and transliterations. [PITH_FULL_IMAGE:figures/full_fig_p013_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

48 extracted references · 43 canonical work pages

  1. [1]

    , author Bell, P

    author Abad, A. , author Bell, P. , author Carmantini, A. , & author Renais, S. ( year 2020 ). title Cross lingual transfer learning for zero-resource domain adaptation . In booktitle ICASSP 2020-2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) \/ (pp. pages 6909--6913 ). organization IEEE

    work page 2020

  2. [2]

    , author Branson, M

    author Ardila, R. , author Branson, M. , author Davis, K. , author Kohler, M. , author Meyer, J. , author Henretty, M. , author Morais, R. , author Saunders, L. , author Tyers, F. , & author Weber, G. ( year 2020 ). title Common voice: A massively-multilingual speech corpus . In booktitle Proceedings of the 12th Language Resources and Evaluation Conferenc...

    work page 2020

  3. [3]

    , author Wang, C

    author Babu, A. , author Wang, C. , author Tjandra, A. , author Lakhotia, K. , author Xu, Q. , author Goyal, N. , author Singh, K. , author von Platen , P. , author Saraf, Y. , author Pino, J. , author Baevski, A. , author Conneau, A. , & author Auli, M. ( year 2022 ). title XLS-R: Self-supervised Cross-lingual Speech Representation Learning at Scale . In...

    doi:10.21437/interspeech.2022-143 2022

  4. [4]

    , author Zhou, Y

    author Baevski, A. , author Zhou, Y. , author Mohamed, A. , & author Auli, M. ( year 2020 ). title wav2vec 2.0: A framework for self-supervised learning of speech representations . journal Advances in neural information processing systems \/ , volume 33 \/ , pages 12449--12460

    work page 2020

  5. [5]

    , author Fainberg, J

    author Bell, P. , author Fainberg, J. , author Klejch, O. , author Li, J. , author Renals, S. , & author Swietojanski, P. ( year 2020 ). title Adaptation algorithms for neural network-based speech recognition: An overview . journal IEEE Open Journal of Signal Processing \/ , volume 2 \/ , pages 33--66

    work page 2020

  6. [6]

    , author Du, J

    author Bu, H. , author Du, J. , author Na, X. , author Wu, B. , & author Zheng, H. ( year 2017 ). title Aishell-1: An open-source mandarin speech corpus and a speech recognition baseline . In booktitle 2017 20th conference of the oriental chapter of the international coordinating committee on speech databases and speech I/O systems and assessment (O-COCOS...

    work page 2017

  7. [7]

    , & author Yeo, E

    author Choi, K. , & author Yeo, E. J. ( year 2022 ). title Opening the black box of wav2vec feature encoder . journal arXiv preprint arXiv:2210.15386 \/ ,

    arXiv 2022

  8. [8]

    , author Baevski, A

    author Conneau, A. , author Baevski, A. , author Collobert, R. , author Mohamed, A. , & author Auli, M. ( year 2021 ). title Unsupervised Cross-Lingual Representation Learning for Speech Recognition . In booktitle Proc. Interspeech 2021 \/ (pp. pages 2426--2430 ). :10.21437/Interspeech.2021-329

    doi:10.21437/interspeech.2021-329 2021

Show all 48 references
  1. [9]

    , author Hasan, S

    author Deselaers, T. , author Hasan, S. , author Bender, O. , & author Ney, H. ( year 2009 ). title A deep learning approach to machine transliteration . In booktitle Proceedings of the Fourth Workshop on Statistical Machine Translation \/ (pp. pages 233--241 )

    2009

  2. [10]

    , & author Lempitsky, V

    author Ganin, Y. , & author Lempitsky, V. ( year 2015 ). title Unsupervised domain adaptation by backpropagation . In booktitle International conference on machine learning \/ (pp. pages 1180--1189 ). organization PMLR

    2015

  3. [11]

    , author Fern \'a ndez, S

    author Graves, A. , author Fern \'a ndez, S. , author Gomez, F. , & author Schmidhuber, J. ( year 2006 ). title Connectionist temporal classification: labelling unsegmented sequence data with recurrent neural networks . In booktitle Proceedings of the 23rd international confer...

    2006

  4. [12]

    , author Moritz, N

    author Higuchi, Y. , author Moritz, N. , author Roux, J. L. , & author Hori, T. ( year 2021 ). title Momentum Pseudo-Labeling for Semi-Supervised Speech Recognition . In booktitle Proc. Interspeech 2021 \/ (pp. pages 726--730 ). :10.21437/Interspeech.2021-571

    2021 doi

  5. [13]

    , author Sriram, A

    author Hsu, W.-N. , author Sriram, A. , author Baevski, A. , author Likhomanenko, T. , author Xu, Q. , author Pratap, V. , author Kahn, J. , author Lee, A. , author Collobert, R. , author Synnaeve, G. , & author Auli, M. ( year 2021 a ). title Robust wav2vec 2.0: Analyzing Dom...

    2021 doi

  6. [14]

    , author Tsai, Y.-H

    author Hsu, W.-N. , author Tsai, Y.-H. H. , author Bolte, B. , author Salakhutdinov, R. , & author Mohamed, A. ( year 2021 b ). title Hubert: How much can a bad teacher benefit asr pre-training? In booktitle ICASSP 2021-2021 IEEE International Conference on Acoustics, Speech a...

    2021

  7. [15]

    , author Li, J

    author Huang, J.-T. , author Li, J. , author Yu, D. , author Deng, L. , & author Gong, Y. ( year 2013 ). title Cross-language knowledge transfer using multilingual deep neural network with shared hidden layers . In booktitle 2013 IEEE international conference on acoustics, spe...

    2013

  8. [16]

    , author Lee, A

    author Kahn, J. , author Lee, A. , & author Hannun, A. ( year 2020 ). title Self-training for end-to-end speech recognition . In booktitle ICASSP 2020-2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) \/ (pp. pages 7084--7088 ). organization IEEE

    2020

  9. [17]

    , author Mittal, A

    author Khare, S. , author Mittal, A. R. , author Diwan, A. , author Sarawagi, S. , author Jyothi, P. , & author Bharadwaj, S. ( year 2021 ). title Low resource asr: The surprising effectiveness of high resource transliteration. In booktitle Proc. Interspeech 2021 \/ (pp. pages...

    2021

  10. [18]

    , author Himeur, Y

    author Kheddar, H. , author Himeur, Y. , author Al-Maadeed, S. , author Amira, A. , & author Bensaali, F. ( year 2023 ). title Deep transfer learning for automatic speech recognition: Towards better generalization . journal Knowledge-Based Systems \/ , volume 277 \/ , pages 110851

    2023

  11. [19]

    , author Moritz, N

    author Khurana, S. , author Moritz, N. , author Hori, T. , & author Le Roux, J. ( year 2021 ). title Unsupervised domain adaptation for speech recognition via uncertainty driven self-training . In booktitle ICASSP 2021-2021 IEEE International Conference on Acoustics, Speech an...

    2021

  12. [20]

    author Kingma, D. P. , & author Ba, J. ( year 2015 ). title Adam: A method for stochastic optimization . In booktitle The Third International Conference on Learning Representations \/

    2015

  13. [21]

    , & author Graehl, J

    author Knight, K. , & author Graehl, J. ( year 1997 ). title Machine transliteration . In booktitle Proceedings of the 35th Annual Meeting of the Association for Computational Linguistics and Eighth Conference of the European Chapter of the Association for Computational Lingui...

    1997

  14. [22]

    , author Pl \'a tek, O

    author Korvas, M. , author Pl \'a tek, O. , author Du s ek, O. , author Z ilka, L. , & author Jurc \' cek, F. ( year 2014 ). title Free english and czech telephone speech corpus shared under the cc-by-sa 3.0 license . In booktitle Proceedings of the Ninth International Confere...

    2014

  15. [23]

    , author Raghunathan, A

    author Kumar, A. , author Raghunathan, A. , author Jones, R. M. , author Ma, T. , & author Liang, P. ( year 2022 ). title Fine-tuning can distort pretrained features and underperform out-of-distribution . In booktitle International Conference on Learning Representations \/

    2022

  16. [24]

    , author Seltzer, M

    author Li, J. , author Seltzer, M. L. , author Wang, X. , author Zhao, R. , & author Gong, Y. ( year 2017 ). title Large-scale domain adaptation via teacher-student learning . journal Proc. Interspeech 2017 \/ , (pp. pages 2386--2390 )

    2017

  17. [25]

    author Li, J. et al. ( year 2022 ). title Recent advances in end-to-end automatic speech recognition . journal APSIPA Transactions on Signal and Information Processing \/ , volume 11 \/

    2022

  18. [26]

    , author Yu, S

    author Li, X. , author Yu, S. , author Lei, Y. , author Li, N. , & author Yang, B. ( year 2023 ). title Intelligent machinery fault diagnosis with event-based camera . journal IEEE Transactions on Industrial Informatics \/ , volume 20 \/ , pages 380--389

    2023

  19. [27]

    , author Long, M

    author Liu, H. , author Long, M. , author Wang, J. , & author Jordan, M. ( year 2019 ). title Transferable adversarial training: A general approach to adapting deep classifiers . In booktitle International Conference on Machine Learning \/ (pp. pages 4013--4022 ). organization PMLR

    2019

  20. [28]

    , & author Hinton, G

    author van der Maaten, L. , & author Hinton, G. ( year 2008 ). title Visualizing data using t-sne . journal Journal of Machine Learning Research \/ , volume 9 \/ , pages 2579--2605

    2008

  21. [29]

    , author Ghahremani, P

    author Manohar, V. , author Ghahremani, P. , author Povey, D. , & author Khudanpur, S. ( year 2018 ). title A teacher-student learning approach for unsupervised domain adaptation of sequence-trained asr models . In booktitle 2018 IEEE Spoken Language Technology Workshop (SLT) ...

    2018

  22. [30]

    , & author Nasersharif, B

    author Naderi, N. , & author Nasersharif, B. ( year 2023 ). title Cross corpus speech emotion recognition using transfer learning and attention-based fusion of wav2vec2 and prosody features . journal Knowledge-Based Systems \/ , volume 277 \/ , pages 110814

    2023

  23. [31]

    , author Edunov, S

    author Ott, M. , author Edunov, S. , author Baevski, A. , author Fan, A. , author Gross, S. , author Ng, N. , author Grangier, D. , & author Auli, M. ( year 2019 ). title fairseq: A fast, extensible toolkit for sequence modeling . In booktitle Proceedings of NAACL-HLT 2019: De...

    2019

  24. [32]

    , author Chen, G

    author Panayotov, V. , author Chen, G. , author Povey, D. , & author Khudanpur, S. ( year 2015 ). title Librispeech: an asr corpus based on public domain audio books . In booktitle 2015 IEEE international conference on acoustics, speech and signal processing (ICASSP) \/ (pp. p...

    2015

  25. [33]

    author Park, D. S. , author Chan, W. , author Zhang, Y. , author Chiu, C.-C. , author Zoph, B. , author Cubuk, E. D. , & author Le, Q. V. ( year 2019 ). title Specaugment: A simple data augmentation method for automatic speech recognition . journal Proc. Interspeech 2019 \/ , ...

    2019

  26. [34]

    , author Chou, J.-C

    author Pasad, A. , author Chou, J.-C. , & author Livescu, K. ( year 2021 ). title Layer-wise analysis of a self-supervised speech representation model . In booktitle 2021 IEEE Automatic Speech Recognition and Understanding Workshop (ASRU) \/ (pp. pages 914--921 ). organization IEEE

    2021

  27. [35]

    , author Wu, Z

    author Peng, K.-C. , author Wu, Z. , & author Ernst, J. ( year 2018 ). title Zero-shot deep domain adaptation . In booktitle Proceedings of the European Conference on Computer Vision (ECCV) \/ (pp. pages 764--781 )

    2018

  28. [36]

    , author Zhang, B

    author Sun, S. , author Zhang, B. , author Xie, L. , & author Zhang, Y. ( year 2017 ). title An unsupervised deep domain adaptation approach for robust speech recognition . journal Neurocomputing \/ , volume 257 \/ , pages 79--87

    2017

  29. [37]

    , author Audhkhasi, K

    author Thomas, S. , author Audhkhasi, K. , & author Kingsbury, B. ( year 2020 ). title Transliteration based data augmentation for training multilingual asr acoustic models in low resource settings. In booktitle Proc. Interspeech 2020 \/ (pp. pages 4736--4740 )

    2020

  30. [38]

    , author Wu, Y

    author Wang, C. , author Wu, Y. , author Qian, Y. , author Kumatani, K. , author Liu, S. , author Wei, F. , author Zeng, M. , & author Huang, X. ( year 2021 ). title Unispeech: Unified speech representation learning with labeled and unlabeled data . In booktitle International ...

    2021

  31. [39]

    , author Dai, Z

    author Wang, Z. , author Dai, Z. , author P \'o czos, B. , & author Carbonell, J. ( year 2019 ). title Characterizing and avoiding negative transfer . In booktitle Proceedings of the IEEE/CVF conference on computer vision and pattern recognition \/ (pp. pages 11293--11302 )

    2019

  32. [40]

    , author Liao, H

    author Wen, G. , author Liao, H. , author Li, H. , author Wen, P. , author Zhang, T. , author Gao, S. , & author Wang, B. ( year 2022 ). title Self-labeling with feature transfer for speech emotion recognition . journal Knowledge-Based Systems \/ , volume 254 \/ , pages 109589

    2022

  33. [41]

    , author Frieske, R

    author Yu, T. , author Frieske, R. , author Xu, P. , author Cahyawijaya, S. , author Yiu, C. T. , author Lovenia, H. , author Dai, W. , author Barezi, E. J. , author Chen, Q. , author Ma, X. et al. ( year 2022 ). title Automatic speech recognition datasets in cantonese: A surv...

    2022

  34. [42]

    , author Chen, M

    author Zhang, S. , author Chen, M. , author Chen, J. , author Li, Y.-F. , author Wu, Y. , author Li, M. , & author Zhu, C. ( year 2021 ). title Combining cross-modal knowledge transfer and semi-supervised learning for speech emotion recognition . journal Knowledge-Based System...

    2021

  35. [43]

    , & author Li, X

    author Zhang, W. , & author Li, X. ( year 2022 ). title Data privacy preserving federated transfer learning in machinery fault diagnostics using prior distributions . journal Structural Health Monitoring \/ , volume 21 \/ , pages 1329--1344

    2022

  36. [44]

    , author Han, W

    author Zhang, Y. , author Han, W. , author Qin, J. , author Wang, Y. , author Bapna, A. , author Chen, Z. , author Chen, N. , author Li, B. , author Axelrod, V. , author Wang, G. et al. ( year 2023 ). title Google usm: Scaling automatic speech recognition beyond 100 languages ...

    2023 arXiv

  37. [45]

    , author Song, Y

    author Zhang, Z.-Q. , author Song, Y. , author Wu, M.-H. , author Fang, X. , author McLoughlin, I. , & author Dai, L.-R. ( year 2022 ). title Cross-lingual self-training to learn multilingual representation for low-resource speech recognition . journal Circuits, Systems, and S...

    2022

  38. [46]

    , author Cheng, G

    author Zhu, H. , author Cheng, G. , author Wang, J. , author Hou, W. , author Zhang, P. , & author Yan, Y. ( year 2023 ). title Boosting cross-domain speech recognition with self-supervision . journal IEEE/ACM Transactions on Audio, Speech, and Language Processing \/ ,

    2023

  39. [47]

    , author Wang, L

    author Zhu, H. , author Wang, L. , author Cheng, G. , author Wang, J. , author Zhang, P. , & author Yan, Y. ( year 2022 ). title Wav2vec-S: Semi-Supervised Pre-Training for Low-Resource ASR . In booktitle Proc. Interspeech 2022 \/ (pp. pages 4870--4874 ). :10.21437/Interspeech...

    2022 doi

  40. [48]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.