Pith. sign in

REVIEW 4 major objections 5 minor 63 references

Breaking the Transcription Bottleneck: Fine-tuning ASR Models for Extremely Low-Resource Fieldwork Languages

T0 review · 4 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 Fine-tuning experiments on five fieldwork languages show MMS beats XLS-R under one hour of transcribed speech, with XLS-R reaching parity once data pass one hour.

desk verdict Useful five-language benchmark of MMS vs XLS-R for fieldwork ASR, but the crossover claim needs variance estimates; worth reviewing with revisions. read the letter →

arxiv 2506.17459 v1 pith:KDK4PJJG submitted 2025-06-20 cs.CL cs.SDeess.AS

Siyu Liang , Gina-Anne Levow This is my paper
classification cs.CLcs.SDeess.AS
keywords automaticspeechrecognitionlow-resourceASRfieldworklanguagesMMSXLS-Rfine-tuningendangeredlanguagedocumentationcharactererrorrate
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

A single hour of fieldwork audio can require up to 50 hours of manual transcription, so choosing an ASR model that works with almost no transcribed data matters for language documentation. The paper tests two fine-tunable multilingual models, MMS and XLS-R, on five typologically varied languages using real, noisy, spontaneous field recordings and controlled training durations of 10, 30, 60, and 120 minutes. Its central claim is a decision rule: MMS gives lower character error rates when less than one hour of transcribed data is available, while XLS-R matches MMS once training data reach about one hour. The paper also shows that both models continue to miss tone, nasality, and vowel-length contrasts, so those categories need targeted treatment beyond model choice.

What carries the argument

The comparison is carried by two wav2vec 2.0-based models adapted in different ways: MMS-1B-l1107, a 1-billion-parameter model with frozen base weights and a 2-million-parameter trainable adapter, and XLS-R-300m, a 300-million-parameter model that is fully fine-tuned. Training is controlled by constructing superset splits of 10, 30, 60, and 120 minutes per language, with a fixed 10-minute test set and Character Error Rate as the primary metric, since CER tracks phoneme-level accuracy better than word error rate for languages without standardized orthographies. This design isolates data duration as the variable that separates MMS's advantage from XLS-R's parity.

What would settle it

Fine-tune MMS and XLS-R on 10, 30, 60, and 120 minutes of a new fieldwork language with a test set of speakers never heard during training; if XLS-R no longer trails below one hour, or if MMS's lead at 10 minutes disappears, the paper's ranking and threshold claim is refuted.

Watch

Extended reading notes

Core claim

With under one hour of transcribed fieldwork speech, fine-tuned MMS achieves lower Character Error Rate than fine-tuned XLS-R across the five test languages; with one hour or more, XLS-R reaches parity. The authors attribute the early advantage to MMS's pre-training on more than a thousand languages and to its built-in ASR fine-tuning plus adapter layers, and XLS-R's catch-up to its more conversationally diverse pre-training corpus. They further show that the advantage is not explained by MMS having seen related Kichwa dialects, since a linear mixed-effects model finds no significant extra gain for Upper Napo Kichwa. Phonologically informed error analysis on Cicipu and Mocho' finds both models struggle with tone, nasality, and consonant/vowel length, with deletions dominating nasality errors and substitutions dominating length errors.

Load-bearing premise

The central claim assumes that a 10-to-120-minute slice of an existing archive behaves like all the data a field linguist would have for a new language, even though the full archive (up to 22 hours) and its speakers were available when the slice was selected.

Editorial extensions

If this is right

  • Field linguists with under an hour of transcribed material should fine-tune MMS first, since it gives lower character error rates under that threshold.
  • The one-hour mark is a practical decision point: beyond it, XLS-R is competitive and may be preferable because its full fine-tuning uses a more conversationally diverse pre-training corpus.
  • Character error rate is the right evaluation target for documentation work, where phonetic accuracy matters more than word-level output.
  • Tone and nasality transcription will not be fixed by model choice alone; targeted augmentation, adapter design, or loss functions are needed for these categories.
  • The observed plateau after roughly one hour suggests that additional transcription effort has diminishing returns for fine-tuning these models.

Reading between the lines

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

  • The one-hour threshold is probably not a universal constant; languages with larger orthographic inventories (Cicipu's 93 characters) or with no related variety in the pretraining data may need more data before XLS-R catches up, and languages closer to pretraining data may need less.
  • Because training and test segments draw on the same speakers, the reported parity point may be optimistic; a held-out-speaker evaluation could push both models' error rates up and possibly change which model leads.
  • A testable extension is continued pre-training on untranscribed field audio: the paper's own reading of MMS's adapter advantage suggests that adding in-language unlabeled data could lower the one-hour threshold for both models.
  • The superset training design means the 120-minute condition contains the 10-minute condition; comparing models under strictly disjoint data draws would more directly test whether the advantage is about quantity or about which utterances are included.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper compares fine-tuned MMS-1B-l1107 and XLS-R-300m on ASR for five low-resource fieldwork languages from ELAR, using controlled amounts of 10, 30, 60, and 120 minutes of training data and a fixed 10-minute test set per language. It reports CER and WER, concludes that MMS is preferable below one hour while XLS-R reaches parity above one hour, and adds phonologically informed error analysis for tone, nasality, and vowel/consonant length in Cicipu and Mocho’. The central quantitative claim is plausible but not yet fully supported: the crossover conclusion rests on single-run point estimates with no variance quantification, and the one-hour threshold is extrapolated from a single 120-minute condition. The paper is a useful empirical contribution for the fieldwork community if these robustness gaps are addressed.

Significance. Strengths of the paper include its use of real fieldwork recordings with environmental noise and spontaneous speech, its coverage of five typologically diverse languages, its choice of CER as the primary metric for languages without standardized orthographies, its candid limitations section, and its attempt to check for a possible Kichwa pre-training advantage with a linear mixed-effects model. The phonological error analysis also goes beyond the usual global error rates and gives field linguists information about tone, nasality, and length contrasts. If the crossover trend is reproducible, the paper would provide a simple, actionable rule for practitioners and a useful benchmark for future low-resource ASR work. However, the headline recommendation is currently based on a small number of point estimates without error bars, confidence intervals, or significance tests, and the abstract/conclusion wording is stronger than the displayed evidence. The practical guidance therefore needs additional statistical support or a more cautious framing before it can be adopted with confidence.

major comments (4)
  1. [Section 4.1, Figure 1; Section 7] The central claim that MMS “consistently achieves lower error rates” below one hour and that XLS-R reaches parity above one hour is based on a single fine-tuning run per model per language per duration (5 languages × 4 durations × 2 models) with no confidence intervals, error bars, or significance tests. Because the test set is a fixed 10-minute sample (Section 2.2) and no held-out speaker is used, the observed per-language differences could be within run-to-run or subset-sampling noise. Section 7 acknowledges shared speakers and the small number of languages but does not mention the absence of variance estimates. Please provide variance estimates (e.g., multiple seeds, bootstrap over test segments, or a significance test) or soften the claim to describe the observed runs; as written, the practical recommendation is not statistically supported.
  2. [Abstract; Section 4.1; Figure 1 caption] The “one-hour threshold” is not directly measured. Training durations are 10, 30, 60, and 120 minutes, and the Figure 1 caption explicitly states that the connecting lines do not imply performance at intermediate durations. The abstract and Section 4.1 nevertheless conclude that “once training data exceed one hour” XLS-R reaches parity, and Section 4.1 further says XLS-R “becomes a more effective option” with approximately one hour or more. The only evidence past one hour is the single 120-minute condition. Please add intermediate points (e.g., 90 minutes) or rephrase the conclusion as “at two hours in these runs” to avoid over-generalizing from one data point.
  3. [Section 2.2; Table 6] The low-resource scenario is simulated by drawing 10–120 minute subsets from archives that contain up to 22.84 hours for Toratán and several cleaned hours for the other languages (Table 6). The paper does not specify how the subsets were selected: random draws, first N utterances, or balanced-by-speaker/genre sampling. If the subsets are random draws from a larger archive, the experimental condition is not equivalent to having only 10 minutes of newly collected fieldwork audio, because the full archive’s speaker and genre coverage is known to the experimenter. This does not invalidate the model comparison, since both models train on identical subsets, but it does affect the external validity of the “extremely low-resource fieldwork” recommendation. Please specify the selection procedure and discuss how this affects the practical guidance.
  4. [Abstract vs. Section 4.1; Figure 1] The abstract says XLS-R “shows parity performance once training data exceed one hour,” while Section 4.1 says XLS-R “becomes a more effective option” when approximately one hour or more is obtainable. Moreover, in Figure 1, MMS appears still lower than XLS-R in several languages at the 120-minute point. The paper should align the abstract, Section 4.1, the conclusion, and the figure, and should avoid claiming a clear advantage for XLS-R beyond one hour when the displayed point estimates do not consistently show it.
minor comments (5)
  1. [Abstract] The phrase “for further provide insights towards practical guidelines” is grammatically incomplete and should be revised to “and provide insights toward practical guidelines.”
  2. [Section 3.2] The fine-tuning procedure is credited only to “von Platen” via a footnote; please add a full bibliographic reference for the MMS adapter recipe.
  3. [Section 4; Figure 1] No numeric CER or WER table is provided in the main text. Adding a compact table of per-language CER at each duration would make the trends in Figure 1 auditable and facilitate comparison with future work.
  4. [Section 4.2] The linear mixed-effects model is underspecified: the text does not define the “time” variable, whether it is log-transformed, or whether random slopes or interactions beyond the reported one were considered. With only five languages, the non-significant interaction (β = 0.021, p = 0.896) should be described as an underpowered check rather than strong evidence of no Kichwa-specific benefit.
  5. [Section 4.3.1; Table 3] Table 3 does not state which training duration the phonological error rates correspond to, whereas Table 4 explicitly reports the XLS-R 120-minute model. If Table 3 is also 120-minute-only, the claim that “both models struggle” with these phonological categories should be restricted to that condition and not generalized across all data sizes.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the comparison is an external benchmark with a fixed held-out test set; the cited prior Mvskoke study is auxiliary, not load-bearing.

full rationale

The paper's central claim (MMS wins below one hour, XLS-R reaches parity near one hour) is an empirical measurement, not a derivation. Section 2.2 fixes a 10-minute test set separate from the 10/30/60/120-minute train+dev subsets, and Figure 1 reports CER/WER on that held-out audio. There is no equation or fitted parameter in the paper that is reused as the predicted quantity: the error rates are computed from model outputs, and the one-hour threshold is a summary of observed curves. The Figure 1 caption explicitly says the connecting lines 'do not imply performance at intermediate durations,' so any over-interpretation of the threshold is a robustness concern rather than circularity. The only self-citations (Mainzinger and Levow 2024; Levow et al. 2021) are not load-bearing: the Mvskoke result is an independent, externally falsifiable study invoked as a possible explanation for MMS's advantage, while the current five-language crossover stands on the paper's own held-out evaluations. The absence of confidence intervals, shared-speaker leakage, and subset-sampling representativeness weaken evidential strength but do not make the comparison reduce to its inputs. No self-definitional, fitted-input-as-prediction, uniqueness-imported, or ansatz-smuggled-by-citation pattern is present.

Assumptions & free parameters 2 free parameters · 3 assumptions · 0 invented entities

The ledger is modest. The paper introduces no new entities or fitted model parameters. The main free choices are the test and development set splits, hyperparameters, and the subset sampling strategy. The central claim depends on these choices and on the reliability of the character-level alignments, so the overall burden is moderate for an empirical benchmark.

free parameters (2)
  • Test set selection (10 minutes per language) = 10 minutes per language
    The fixed 10-minute test set is a choice made by the authors. Performance numbers and the crossover claim depend on this particular test split, and no confidence intervals are reported.
  • Hyperparameters: learning rate MMS 1e-3, XLS-R 3e-4, max epochs 30, early stopping patience 3 = MMS 1e-3, XLS-R 3e-4, 30 epochs, patience 3
    These are chosen by hand from the cited blog tutorial and not tuned per language. Different hyperparameters could shift the crossover, so they function as a configuration choice that affects the central comparison.
assumptions (3)
  • domain assumption The 10-120 minute subsets are representative of the full archive distribution for each language.
    Section 2.2 explains that subsets are drawn from the cleaned archive and that larger splits are supersets of smaller ones, but the paper does not control for genre, speaker, or recording condition balance across durations. The recommendation depends on this representativeness assumption.
  • domain assumption Character-level alignment is a reliable way to assign errors to phonological categories.
    The error analysis in Section 4.3 and Appendix C computes substitution, deletion, and insertion counts from character alignments, but the alignment method and its validation are only summarized in Appendix C. If alignments are inaccurate, the category-specific error rates would be affected.
  • domain assumption CER is the appropriate optimization and evaluation metric for fieldwork ASR.
    The authors choose CER as the training and evaluation metric in Section 3.2 and 4 because field linguists lack word-level orthographic resources. This is a reasonable domain choice, but it also shapes the result because MMS and XLS-R are compared on character accuracy, not on the practical usability of transcripts.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Breaking the Transcription Bottleneck: Fine-tuning ASR Models for Extremely Low-Resource Fieldwork Languages." pith.science (2026). https://pith.science/paper/KDK4PJJG

@misc{pith2026250617459,
  author       = {Pith},
  title        = {Pith review of: Breaking the Transcription Bottleneck: Fine-tuning ASR Models for Extremely Low-Resource Fieldwork Languages},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KDK4PJJG}},
  note         = {Machine review of arXiv:2506.17459}
}
read the original abstract

Automatic Speech Recognition (ASR) has reached impressive accuracy for high-resource languages, yet its utility in linguistic fieldwork remains limited. Recordings collected in fieldwork contexts present unique challenges, including spontaneous speech, environmental noise, and severely constrained datasets from under-documented languages. In this paper, we benchmark the performance of two fine-tuned multilingual ASR models, MMS and XLS-R, on five typologically diverse low-resource languages with control of training data duration. Our findings show that MMS is best suited when extremely small amounts of training data are available, whereas XLS-R shows parity performance once training data exceed one hour. We provide linguistically grounded analysis for further provide insights towards practical guidelines for field linguists, highlighting reproducible ASR adaptation approaches to mitigate the transcription bottleneck in language documentation.

Figures

Figures reproduced from arXiv: 2506.17459 by the authors.

Figure 1
Figure 1. CER comparison for MMS-1b1107 and XLS￾R-300m models across five languages. The MMS model performs markedly better under extremely low-resource settings (less than 1 hour), but XLS-R performs simi￾larly well with 2 hours of data. Points are connected to aid trend reading and do not imply performance at intermediate durations [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. WER comparison for MMS-1b1107 and XLS￾R-300m models across five languages. mization strategy, and training details are provided in Appendix A. 4 Results Overall, the performance of the MMS and XLS￾R models in automatic speech recognition (ASR) tasks on low-resource language data is comparable, though nuanced differences emerge depending on the availability of training data. In general, the MMS model outperforms XLS-… view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

63 extracted references · 50 canonical work pages

  1. [1]

    Antonis Anastasopoulos and David Chiang. 2018. https://doi.org/10.48550/arXiv.1803.08991 Leveraging translations for speech transcription in low-resource settings . arXiv preprint. ArXiv:1803.08991 [cs]

    work page Pith review arXiv doi:10.48550/arxiv.1803.08991 2018

  2. [2]

    Ahmed Adel Attia, Dorottya Demszky, Tolulope Ogunremi, Jing Liu, and Carol Espy-Wilson. 2024. https://doi.org/10.48550/arXiv.2405.13018 Continued Pretraining for Domain Adaptation of Wav2vec2 .0 in Automatic Speech Recognition for Elementary Math Classroom Settings . arXiv preprint. ArXiv:2405.13018 [cs] version: 1

    work page Pith review arXiv doi:10.48550/arxiv.2405.13018 2024

  3. [3]

    Austin and Julia Sallabank

    Peter K. Austin and Julia Sallabank. 2011. The Cambridge Handbook of Endangered Languages . Cambridge University Press. Google-Books-ID: 0XZRauYgO6AC

    work page 2011

  4. [4]

    Arun Babu, Changhan Wang, Andros Tjandra, Kushal Lakhotia, Qiantong Xu, Naman Goyal, Kritika Singh, Patrick von Platen, Yatharth Saraf, Juan Pino, Alexei Baevski, Alexis Conneau, and Michael Auli. 2021. https://doi.org/10.48550/arXiv.2111.09296 XLS - R : Self -supervised Cross -lingual Speech Representation Learning at Scale . arXiv preprint. ArXiv:2111.0...

    arXiv doi:10.48550/arxiv.2111.09296 2021

  5. [5]

    Alexei Baevski, Wei-Ning Hsu, Alexis CONNEAU, and Michael Auli. 2021. https://proceedings.neurips.cc/paper/2021/hash/ea159dc9788ffac311592613b7f71fbb-Abstract.html Unsupervised Speech Recognition . In Advances in Neural Information Processing Systems , volume 34, pages 27826--27839. Curran Associates, Inc

    work page 2021

  6. [6]

    Alexei Baevski, Yuhao Zhou, Abdelrahman Mohamed, and Michael Auli. 2020. https://proceedings.neurips.cc/paper/2020/hash/92d1e1eb1cd6f9fba3227870bb6d7f07-Abstract.html wav2vec 2.0: A Framework for Self - Supervised Learning of Speech Representations . In Advances in Neural Information Processing Systems , volume 33, pages 12449--12460. Curran Associates, Inc

    work page 2020

  7. [7]

    Sainath, and Trevor Strohman

    Junwen Bai, Bo Li, Qiujia Li, Tara N. Sainath, and Trevor Strohman. 2024. https://doi.org/10.1109/ICASSP48485.2024.10447399 Efficient Adapter Finetuning for Tail Languages in Streaming Multilingual ASR . In ICASSP 2024 - 2024 IEEE International Conference on Acoustics , Speech and Signal Processing ( ICASSP ) , pages 10841--10845. ISSN: 2379-190X

    arXiv 2024

  8. [8]

    Russell Barlow. 2018 a . http://hdl.handle.net/2196/00-0000-0000-000F-CB61-A Documentation of Ulwa , an endangered language of Papua New Guinea

    work page 2018

Show all 63 references
  1. [9]

    Russell Barlow. 2018 b . http://hdl.handle.net/10125/62506 A Grammar of Ulwa

    2018

  2. [10]

    Martijn Bartelds, Nay San, Bradley McDonnell, Dan Jurafsky, and Martijn Wieling. 2023. https://doi.org/10.48550/arXiv.2305.10951 Making More of Little Data : Improving Low - Resource Automatic Speech Recognition Using Data Augmentation . arXiv preprint. ArXiv:2305.10951 [cs]

    2023 doi arXiv

  3. [11]

    Steven Bird. 2020. https://doi.org/10.1162/coli_a_00387 Sparse Transcription . Computational Linguistics, 46(4):713--744

    2020 doi

  4. [12]

    Hennie Brugman and Albert Russel. 2004. Annotating Multi -media / Multi -modal resources with ELAN

    2004

  5. [13]

    Xuankai Chang, Takashi Maekaku, Yuya Fujita, and Shinji Watanabe. 2022. https://doi.org/10.48550/arXiv.2204.00540 End-to- End Integration of Speech Recognition , Speech Enhancement , and Self - Supervised Learning Representation . arXiv preprint. ArXiv:2204.00540 [cs]

    2022 doi arXiv

  6. [14]

    Rolando Coto-Solano. 2021. https://doi.org/10.18653/v1/2021.americasnlp-1.20 Explicit Tone Transcription Improves ASR Performance in Extremely Low - Resource Languages : A Case Study in Bribri . In Proceedings of the First Workshop on Natural Language Processing for Indigenous...

    2021 doi

  7. [15]

    Xiaodong Cui, Brian Kingsbury, Jia Cui, Bhuvana Ramabhadran, Andrew Rosenberg, Mohammad Sadegh Rasooli, Owen Rambow, Nizar Habash, and Vaibhava Goel. 2014. https://doi.org/10.21437/Interspeech.2014-477 Improving deep neural network acoustic modeling for audio corpus indexing u...

    2014 doi

  8. [16]

    Mitchell DeHaven and Jayadev Billa. 2022. https://doi.org/10.48550/arXiv.2207.00659 Improving Low - Resource Speech Recognition with Pretrained Speech Models : Continued Pretraining vs. Semi - Supervised Training . arXiv preprint. ArXiv:2207.00659 [cs]

    2022 doi arXiv

  9. [17]

    Black, Laurent Besacier, Sakriani Sakti, and Emmanuel Dupoux

    Ewan Dunbar, Robin Algayres, Julien Karadayi, Mathieu Bernard, Juan Benjumea, Xuan-Nga Cao, Lucie Miskic, Charlotte Dugrain, Lucas Ondel, Alan W. Black, Laurent Besacier, Sakriani Sakti, and Emmanuel Dupoux. 2019. https://doi.org/10.48550/arXiv.1904.11469 The Zero Resource Spe...

    2019 doi arXiv

  10. [18]

    Eberhard, Gary F

    David M. Eberhard, Gary F. Simons, and Charles D. Fennig. 2024. https://www.ethnologue.com/language/qvo Napo Quichua . Edition: 27 Publisher: SIL International

    2024

  11. [19]

    Eberhard, Gary F

    David M. Eberhard, Gary F. Simons, and Charles D. Fennig. 2025. https://www.ethnologue.com/language/quw Tena Lowland Quichua . Edition: 26 Publisher: SIL International

    2025

  12. [20]

    Godfrey, E.C

    J.J. Godfrey, E.C. Holliman, and J. McDaniel. 1992. https://doi.org/10.1109/ICASSP.1992.225858 SWITCHBOARD : telephone speech corpus for research and development . In [ Proceedings ] ICASSP -92: 1992 IEEE International Conference on Acoustics , Speech , and Signal Processing ,...

    1992

  13. [21]

    Karolina Grzech. 2020. http://hdl.handle.net/2196/00-0000-0000-000C-F5FB-A Upper Napo Kichwa : a documentation of linguistic and cultural practices

    2020

  14. [22]

    Frantisek Grézl, Martin Karafiát, and Karel Veselý. 2014. https://doi.org/10.1109/ICASSP.2014.6855089 Adaptation of multilingual stacked bottle-neck neural network structure for new language . In 2014 IEEE International Conference on Acoustics , Speech and Signal Processing ( ...

    2014

  15. [23]

    Séverine Guillaume, Guillaume Wisniewski, Benjamin Galliot, Minh-Châu Nguyen, Maxime Fily, Guillaume Jacques, and Alexis Michaud. 2022 a . https://doi.org/10.21437/Interspeech.2022-11314 Plugging a neural phoneme recognizer into a simple language model: a workflow for low-reso...

    2022 doi

  16. [24]

    Séverine Guillaume, Guillaume Wisniewski, Cécile Macaire, Guillaume Jacques, Alexis Michaud, Benjamin Galliot, Maximin Coavoux, Solange Rossato, Minh-Châu Nguyen, and Maxime Fily. 2022 b . https://doi.org/10.18653/v1/2022.computel-1.21 Fine-tuning pre-trained models for Automa...

    2022 doi

  17. [25]

    Himmelmann

    Nikolaus P. Himmelmann. 1998. https://doi.org/10.1515/ling.1998.36.1.161 Documentary and descriptive linguistics . 36(1):161--196. Publisher: De Gruyter Mouton Section: Linguistics

    1998 doi

  18. [26]

    Nikolaus P Himmelmann and John U Wolff. 1999. Toratán ( Ratahan ) , volume 130. Lincom Europa

    1999

  19. [27]

    Neil Houlsby, Andrei Giurgiu, Stanislaw Jastrzebski, Bruna Morrone, Quentin De Laroussilhe, Andrea Gesmundo, Mona Attariyan, and Sylvain Gelly. 2019. https://proceedings.mlr.press/v97/houlsby19a.html Parameter- Efficient Transfer Learning for NLP . In Proceedings of the 36th I...

    2019

  20. [28]

    Wei-Ning Hsu, Benjamin Bolte, Yao-Hung Hubert Tsai, Kushal Lakhotia, Ruslan Salakhutdinov, and Abdelrahman Mohamed. 2021. https://doi.org/10.1109/TASLP.2021.3122291 HuBERT : Self - Supervised Speech Representation Learning by Masked Prediction of Hidden Units . IEEE/ACM Transa...

    2021

  21. [29]

    Robert Jimerson, Zoey Liu, and Emily Prud'hommeaux. 2023. https://doi.org/10.18653/v1/2023.acl-short.87 An (unhelpful) guide to selecting the best ASR architecture for your under-resourced language . In Proceedings of the 61st Annual Meeting of the Association for Computationa...

    2023 doi

  22. [30]

    Austin Jones, Shulin Zhang, John Hale, Margaret Renwick, Zvjezdana Vrzic, and Keith Langston. 2024. https://doi.org/10.18653/v1/2024.fieldmatters-1.8 Comparing Kaldi - Based Pipeline Elpis and Whisper for Čakavian Transcription . In Proceedings of the 3rd Workshop on NLP Appli...

    2024 doi

  23. [31]

    Anthony Jukes. 2010. http://hdl.handle.net/2196/00-0000-0000-0001-56FB-9 Documentation of Toratán ( Ratahan )

    2010

  24. [32]

    Shreya Khare, Ashish Mittal, Anuj Diwan, Sunita Sarawagi, Preethi Jyothi, and Samarth Bharadwaj. 2021. https://doi.org/10.21437/Interspeech.2021-2062 Low Resource ASR : The Surprising Effectiveness of High Resource Transliteration . In Interspeech 2021, pages 1529--1533. ISCA

    2021 doi

  25. [33]

    Jiyeon Kim, Mehul Kumar, Dhananjaya Gowda, Abhinav Garg, and Chanwoo Kim. 2021. https://doi.org/10.48550/arXiv.2111.10047 Semi-supervised transfer learning for language expansion of end-to-end speech recognition models to low-resource languages . arXiv preprint. ArXiv:2111.100...

    2021 doi arXiv

  26. [34]

    Ahn, and Emily M

    Gina-Anne Levow, Emily P. Ahn, and Emily M. Bender. 2021. https://doi.org/10.33011/computel.v1i.967 Developing a Shared Task for Speech Processing on Endangered Languages . Proceedings of the Workshop on Computational Methods for Endangered Languages, 1:96--106

    2021 doi

  27. [35]

    Zoey Liu, Justin Spence, and Emily Prud'hommeaux. 2023. https://doi.org/10.18653/v1/2023.eacl-main.10 Investigating data partitioning strategies for crosslinguistic low-resource ASR evaluation . In Proceedings of the 17th Conference of the European Chapter of the Association f...

    2023 doi

  28. [36]

    Julia Mainzinger and Gina-Anne Levow. 2024. https://doi.org/10.18653/v1/2024.acl-srw.16 Fine- Tuning ASR models for Very Low - Resource Languages : A Study on Mvskoke . In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics ( Volume 4: Stude...

    2024 doi

  29. [37]

    Stuart McGill. 2012. http://hdl.handle.net/2196/00-0000-0000-0001-7D83-D Cicipu documentation

    2012

  30. [38]

    Stuart McGill. 2014. https://doi.org/10.1017/S002510031400022X Cicipu . Journal of the International Phonetic Association, 44(3):303--318

    2014 doi

  31. [39]

    Yajie Miao, Florian Metze, and Shourabh Rawat. 2013. https://doi.org/10.1109/ASRU.2013.6707763 Deep maxout networks for low-resource speech recognition . In 2013 IEEE Workshop on Automatic Speech Recognition and Understanding , pages 398--403

    2013

  32. [40]

    Benjamin Milde and Arne Köhn. 2018. https://doi.org/10.48550/arXiv.1807.10311 Open Source Automatic Speech Recognition for German . arXiv preprint. ArXiv:1807.10311 [cs]

    2018 doi arXiv

  33. [41]

    Karol Nowakowski, Michal Ptaszynski, Kyoko Murasaki, and Jagna Nieuważny. 2023. https://doi.org/10.1016/j.ipm.2022.103148 Adapting Multilingual Speech Representation Model for a New , Underresourced Language through Multilingual Fine -tuning and Continued Pretraining . Informa...

    2023

  34. [42]

    Erin O’Rourke and Tod D. Swanson. 2013. https://www.jstor.org/stable/26351942 Tena Quichua . Journal of the International Phonetic Association, 43(1):107--120. Publisher: Cambridge University Press

    2013

  35. [43]

    Naomi Elizabeth Palosaari. 2011. https://www.proquest.com/docview/863689180/abstract/E7686087339D442BPQ/1 Topics in Mocho ' phonology and morphology . Ph. D ., The University of Utah, United States -- Utah. ISBN: 9781124576213

    2011

  36. [44]

    Vassil Panayotov, Guoguo Chen, Daniel Povey, and Sanjeev Khudanpur. 2015. https://doi.org/10.1109/ICASSP.2015.7178964 Librispeech: An ASR corpus based on public domain audio books . In 2015 IEEE International Conference on Acoustics , Speech and Signal Processing ( ICASSP ) , ...

    2015

  37. [45]

    Pillai, Kavya Manohar, Basil K

    Leena G. Pillai, Kavya Manohar, Basil K. Raju, and Elizabeth Sherly. 2024. https://doi.org/10.48550/arXiv.2411.04573 Multistage Fine -tuning Strategies for Automatic Speech Recognition in Low -resource Languages . arXiv preprint. ArXiv:2411.04573 [cs]

    2024 doi arXiv

  38. [46]

    Daniel Povey, Arnab Ghoshal, Gilles Boulianne, Lukas Burget, Ondrej Glembek, Nagendra Goel, Mirko Hannemann, Petr Motlicek, Yanmin Qian, Petr Schwarz, Jan Silovsky, Georg Stemmer, and Karel Vesely. 2011. The Kaldi Speech Recognition Toolkit . In IEEE 2011 workshop on automatic...

    2011

  39. [47]

    Vineel Pratap, Andros Tjandra, Bowen Shi, Paden Tomasello, Arun Babu, Sayani Kundu, Ali Elkahky, Zhaoheng Ni, Apoorv Vyas, Maryam Fazel-Zarandi, Alexei Baevski, Yossi Adi, Xiaohui Zhang, Wei-Ning Hsu, Alexis Conneau, and Michael Auli. 2024. http://jmlr.org/papers/v25/23-1318.h...

    2024

  40. [48]

    Agnieszka Przezdziak. 2024. https://urn.kb.se/resolve?urn=urn:nbn:se:uu:diva-533377 Optimizing Speech Recognition for Low - Resource Languages : Northern Sotho

    2024

  41. [49]

    Jaime Pérez González. 2018. http://hdl.handle.net/2196/00-0000-0000-0010-7985-9 Documentation of Mocho ’ ( Mayan ): Language Preservation through Community Awareness and Engagement

    2018

  42. [50]

    Alec Radford, Jong Wook Kim, Tao Xu, Greg Brockman, Christine Mcleavey, and Ilya Sutskever. 2023. https://proceedings.mlr.press/v202/radford23a.html Robust Speech Recognition via Large - Scale Weak Supervision . In Proceedings of the 40th International Conference on Machine Le...

    2023

  43. [51]

    Sanjay Rijal, Shital Adhikari, Manish Dahal, Manish Awale, and Vaghawan Ojha. 2024. https://doi.org/10.48550/arXiv.2411.12587 Whisper Finetuning on Nepali Language . arXiv preprint. ArXiv:2411.12587 [cs]

    2024 doi arXiv

  44. [52]

    Amith, Rey Castillo García, Esteban Guadalupe Sierra, Kevin Duh, and Shinji Watanabe

    Jiatong Shi, Jonathan D. Amith, Rey Castillo García, Esteban Guadalupe Sierra, Kevin Duh, and Shinji Watanabe. 2021. https://doi.org/10.48550/arXiv.2101.10877 Leveraging End -to- End ASR for Endangered Language Documentation : An Empirical Study on Yoloxóchitl Mixtec . arXiv p...

    2021 doi arXiv

  45. [53]

    Claytone Sikasote and Antonios Anastasopoulos. 2021. https://doi.org/10.48550/arXiv.2102.04889 BembaSpeech : A Speech Recognition Corpus for the Bemba Language . arXiv preprint. ArXiv:2102.04889 [cs]

    2021 doi arXiv

  46. [54]

    Bao Thai, Robert Jimerson, Raymond Ptucha, and Emily Prud'hommeaux. 2020. https://aclanthology.org/2020.sltu-1.17/ Fully Convolutional ASR for Less - Resourced Endangered Languages . In Proceedings of the 1st Joint Workshop on Spoken Language Technologies for Under -resourced ...

    2020

  47. [55]

    Nick Thieberger. 2012. The Oxford Handbook of Linguistic Fieldwork . OUP Oxford. Google-Books-ID: 86AE2\_0nPbkC

    2012

  48. [56]

    Geoffroy Vanderreydt, François Remy, and Kris Demuynck. 2022. https://doi.org/10.21437/Interspeech.2022-10744 Transfer Learning from Multi - Lingual Speech Translation Benefits Low - Resource Speech Recognition . In Interspeech 2022, pages 3053--3057. ISCA

    2022 doi

  49. [57]

    Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, Anthony Moi, Pierric Cistac, Tim Rault, Remi Louf, Morgan Funtowicz, Joe Davison, Sam Shleifer, Patrick von Platen, Clara Ma, Yacine Jernite, Julien Plu, Canwen Xu, Teven Le Scao, Sylvain Gugger, Mari...

    2020 doi

  50. [58]

    Michael Wroblewski. 2012. https://doi.org/10.1111/j.1548-1395.2012.01134.x Amazonian Kichwa Proper : Ethnolinguistic Domain in Pan - Indian Ecuador . Journal of Linguistic Anthropology, 22(1):64--86. \_eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1111/j.1548-1395.2012.01134.x

    2012

  51. [59]

    Thimmaraja Yadava and H S Jayanna

    G. Thimmaraja Yadava and H S Jayanna. 2017. https://doi.org/10.1109/ICACCI.2017.8126111 Development and comparison of ASR models using kaldi for noisy and enhanced kannada speech data . 2017 International Conference on Advances in Computing, Communications and Informatics (ICA...

    2017

  52. [60]

    Binbin Zhang, Hang Lv, Pengcheng Guo, Qijie Shao, Chao Yang, Lei Xie, Xin Xu, Hui Bu, Xiaoyu Chen, Chenchen Zeng, Di Wu, and Zhendong Peng. 2022. https://doi.org/10.48550/arXiv.2110.03370 WenetSpeech : A 10000+ Hours Multi -domain Mandarin Corpus for Speech Recognition . arXiv...

    2022 doi arXiv

  53. [61]

    Xiaohui Zhang, Jan Trmal, Daniel Povey, and Sanjeev Khudanpur. 2014. https://doi.org/10.1109/ICASSP.2014.6853589 Improving deep neural network acoustic models using generalized maxout networks . In 2014 IEEE International Conference on Acoustics , Speech and Signal Processing ...

    2014

  54. [62]

    online" 'onlinestring :=

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint eprinttype howpublished institution journal key month note number organization pages publisher school series title type volume year doi pubmed url lastchecked label extra.label sort.label short.list...

  55. [63]

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