Pith. sign in

REVIEW 3 major objections 3 minor 1 cited by

Benchmarking Training Paradigms, Dataset Composition, and Model Scaling for Child ASR in ESPnet

T0 review · 3 major / 3 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read The paper claims that self-supervised speech models trained chiefly on adult audio embed an adult bias that hurts child ASR, that flat-start training on pooled child corpora mitigates it, and that zero-shot model scaling for child speech pl

desk verdict A genuinely useful, reproducible child ASR benchmark whose headline claim about adult bias in SSL features is plausible but not established by the experiments as run. read the letter →

arxiv 2508.16576 v1 pith:QXU6MSR5 submitted 2025-08-22 cs.LG

Anyu Ying , Natarajan Balaji Shankar , Chyi-Jiunn Lin , Mohan Shi , Pu Wang , Hye-jin Shim , Siddhant Arora , Hugo Van hamme
show 2 more authors
Abeer Alwan Shinji Watanabe
This is my paper
classification cs.LG
keywords childautomaticspeechrecognitionflat-starttrainingself-supervisedrepresentationbiasadult-speechmodelscalingplateaumulti-corpusspeakerverificationfoundationmodels
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper asks how much of child-speech recognition difficulty is a training-paradigm problem rather than an acoustics problem, and answers with a controlled comparison: fine-tuning adult-trained models against flat-start training run from scratch, inside one open speech-processing toolkit, across three child corpora and multiple front-ends and decoders. Its central claim is that self-supervised speech representations (WavLM, XEUS) are biased toward adult speech: in multi-corpus flat-start training, these SSL front-ends fail to beat plain filterbank features, the opposite of what holds in adult ASR, and the paper interprets that gap as inherited adult bias that flat-start training on child speech partially corrects. It also finds that zero-shot model scaling for child ASR improves consistently up to about one billion parameters and then plateaus or degrades, across both open (OWLS) and closed (Whisper) model families. If these claims hold, practitioners get a concrete recipe — pooled child corpora, flat-start training, filterbank features, models near 1B parameters — and a warning that bigger or more foundational is not automatically better for children's voices.

What carries the argument

The controlled comparison grid itself is the engine: a 12-layer E-Branchformer encoder as a fixed backbone, paired with three decoding strategies (attention encoder-decoder, RNN transducer, and CTC with or without an external language model), fed by one of three front-ends (filterbank, WavLM, XEUS), trained on a single corpus or on all three pooled, with continuous-versus-discrete (K-means) feature variants as an extra axis. The load-bearing comparison is the multi-corpus flat-start table, where SSL-front-end rows are read against filterbank rows to support the adult-bias claim. The scaling claim rides on zero-shot WER-versus-parameter curves for the Whisper and OWLS model families, with mod

What would settle it

Two checks would settle the claims. (1) Re-run the multi-corpus flat-start grid with WavLM/XEUS front-ends adapted on child speech: if adapted SSL features then beat filterbanks across all four corpora, the adult-bias attribution is supported; if they still lag, the bias is not in the representations alone. (2) Fine-tune the 1B and 9B OWLS models on identical pooled child data: if 9B beats 1B by more than noise, the ~1B plateau is an artifact of zero-shot evaluation rather than a true scaling ceiling.

Watch

Extended reading notes

Core claim

Comparing flat-start training and fine-tuning across three child corpora (MyST, OGI Kids, CMU Kids), the paper reports that SSL-front-end models using WavLM or XEUS representations lose their usual advantage over filterbanks once training moves to a pooled multi-corpus setup, sometimes clearly (OGI Script: 7.4–10.4 WER vs 2.2–2.6 for filterbanks). The authors read this as evidence that SSL representations are biased toward adult speech, since these models were pre-trained almost exclusively on adult audio, and argue that flat-start training on child speech is the mitigation because it learns child-specific patterns without inheriting those biases. Zero-shot scaling curves of Whisper (39M–1.5

Load-bearing premise

The adult-bias verdict assumes that a frozen SSL feature extractor fed into a fixed 12-layer E-Branchformer decoder fairly reveals representation quality: if the SSL models had been fine-tuned or the decoder had interacted differently with feature statistics, the WavLM/XEUS shortfalls could reflect adaptation procedure rather than adult bias. The 1B scaling plateau, meanwhile, is measured in zero-shot evaluation and assumes zero-shot WER stands in for how model size behaves u

Editorial extensions

If this is right

  • In multi-corpus child ASR, plain filterbank features can match or beat frozen SSL features, so SSL front-ends should not be assumed beneficial for child speech without child-adapted pre-training.
  • Zero-shot model scaling for child ASR stops paying off near 1B parameters, so compute beyond that point should go into training data, upsampling, or paradigm choice instead of raw model size.
  • Flat-start training on pooled child corpora, with upsampling of underrepresented age groups, is a workable alternative to fine-tuning adult models when avoiding inherited adult biases matters.
  • Continuous SSL features beat K-means discrete tokens on child speech, so discretization at current cluster granularity is a lossy choice for this domain.
  • Whisper's age-inconsistent results warn that closed-training-data models are unreliable instruments for child-speech analysis; open-data models give interpretable, reproducible curves.

Reading between the lines

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

  • The adult-bias claim implies a test the paper does not run: adapt WavLM/XEUS on child speech and repeat the multi-corpus flat-start grid — if the SSL advantage over filterbanks reappears, the bias explanation is confirmed; if not, the decoder or data mix shares the blame.
  • The ~1B plateau is measured zero-shot; fine-tuned larger models might still scale further, so the ceiling is established for zero-shot transfer, not necessarily for model capacity on child speech.
  • The three corpora use different transcription conventions (verbatim disfluency-preserving versus intended-target text), which likely inflates multi-corpus WERs; normalizing annotation style could shrink the apparent cross-corpus difficulty.
  • The speaker-verification age gradient (EER 13.2 for ages 4–7 versus 3.7 for ages 11–15) suggests the child-speech brittleness is embedding-level, so other downstream tasks such as translation, diarization, or keyword spotting likely inherit the same age asymmetry.
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 / 3 minor

Summary. The paper benchmarks child ASR in a unified ESPnet framework, comparing flat-start training against fine-tuning of pre-trained ASR models across three child speech corpora (MyST, OGI Script/Spon, CMU Kids). It evaluates Fbank versus frozen SSL front-ends (WavLM, XEUS), continuous versus discrete representations, single- versus multi-corpus training, and model scaling via zero-shot evaluation of Whisper and OWLS models. The authors claim that SSL representations are biased toward adult speech, that flat-start training on child speech mitigates this bias, that model scaling improves WER up to about 1B parameters and then plateaus, and that age-based analysis underscores the limitations of proprietary models like Whisper. The paper provides public ESPnet recipes and positions itself as a reproducible benchmark.

Significance. If the headline claims are established, this would be a valuable benchmark for child ASR: the comparison of flat-start versus fine-tuning across multiple corpora and decoder architectures is not otherwise systematically documented, and the public ESPnet recipes with detailed data preparation are a concrete reproducibility contribution. The emphasis on open-data models and the age-grouped analysis of ASR and speaker verification are also useful for the community. However, the two central interpretive claims—the adult-bias attribution of SSL representations and the scaling plateau—are not fully supported by the evidence as presented. The adult-bias conclusion relies on frozen-feature probes with no fine-tuned or adapted SSL condition, and the scaling conclusion relies on zero-shot evaluation where model size is confounded with training data and architecture. The benchmark itself is solid and useful, but the load-bearing interpretations need additional experiments or careful re-scoping.

major comments (3)
  1. [§4.2, Table 4, rows 1–6] The claim that SSL representations are 'biased toward adult speech' is underdetermined. The evidence compares Fbank with WavLM/XEUS used as fixed feature extractors feeding a randomly initialized 12-layer E-Branchformer AED. This setup conflates representation quality with feature–decoder mismatch: frozen SSL features have different scale, dimensionality, and statistics than Fbank, and no adapter, normalization, or fine-tuning is reported. The results are also mixed: in multi-corpus training (Table 4), WavLM outperforms Fbank on OGI Spon (21.3/20.9 vs 24.5/24.2) and CMU Kids (10.9/11.8 vs 15.5/16.4), while Fbank is better on OGI Script (2.2/2.6 vs 7.4/10.4) and comparable on MyST. A 'bias toward adult speech' should predict systematic degradation, not corpus-dependent outcomes. Moreover, the statement that 'flat-start training on child speech mitigates these biases' is not directly teste
  2. [§4.5, Figure 1, Table 5] The scaling claim—'consistent improvements up to 1B parameters, beyond which performance plateaus'—is based solely on zero-shot evaluation of Whisper and OWLS models. This is not a controlled scaling experiment: model family, training data size, architecture, and decoding strategy all vary alongside parameter count. The paper does not report fine-tuned WER as a function of model size, so the conclusion cannot be extended to child ASR systems that are actually trained or adapted. The tabulated age-group results in Table 5 also do not show the claimed monotonic improvement: OWLS 0.5B gives 57.2/23.6/23.2, while OWLS 1B gives 75.2/24.7/18.9, a large degradation on ages 4–7 and 8–10. Please either restrict the claim to zero-shot evaluation, report fine-tuning scaling, or provide a per-corpus plot with confidence intervals that supports the 'consistent' wording.
  3. [§4.2–§4.5, Tables 2, 4, 5] Several headline conclusions rest on WER differences that are within a fraction of a point or on a single split, and only a few cells in Table 2 are marked as statistically significant. For example, in Table 4 the MyST test difference between Fbank AED (13.5) and XEUS (13.2) is 0.3 points, and the CMU Kids difference between Fbank (16.4) and XEUS (16.0) is 0.4 points. Without confidence intervals, significance tests, or at least multiple-seed variation, statements such as 'the performance gap narrows' and 'flat-start Fbank models show notable improvements' are not quantitatively grounded. Please report error bars or significance testing for the key comparisons, especially where the direction of the effect is used to support the adult-bias and multi-corpus conclusions.
minor comments (3)
  1. [§3.1, Table 1] The MyST corpus is filtered using Whisper-largeV2, and the paper later evaluates Whisper models on the same corpus (Tables 2, 5, Figure 1). This creates a potential feedback loop: the filter may preferentially select utterances that Whisper transcribes well, which could bias comparisons between Whisper and other models. The paper should acknowledge this and, ideally, quantify the effect by comparing filtering with an open-source model or reporting the fraction of utterances removed per dataset.
  2. [§4.6, Table 5] Table 5 does not state which training set (single-corpus or multi-corpus, with or without upsampling) is used for the flat-start and fine-tuned rows. Since Section 4.4 shows that multi-corpus training and upsampling materially change WER, the age-group comparison is hard to interpret without this information. Please specify the training setup for each row.
  3. [§2.5, Figure 1] Whisper large-v3 is excluded 'due to differences in training data composition compared to smaller Whisper models,' but the specific composition differences are not described. Please give a one-sentence explanation or citation. Also, the caption of Figure 1 would benefit from a note on whether the models are evaluated in the same decoding condition (e.g., greedy vs beam search).

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; the paper is an empirical benchmark with direct evaluations, and the adult-bias attribution, while underdetermined, is not a circular reduction.

full rationale

This is an empirical benchmark paper, not a derivational or predictive modeling paper. No parameter is fitted to a subset of data and then renamed as a prediction; the upsample ratios, cluster counts, and model sizes are set a priori and evaluated directly. The central claim that SSL representations are biased toward adult speech is inferred from Table 4, where frozen WavLM/XEUS features sometimes underperform Fbank in multi-corpus flat-start training. This inference is open to alternative explanations (feature–decoder mismatch, data scarcity, optimization dynamics), but that is an internal-validity concern, not a circularity: the paper does not define 'adult bias' in terms of the observed WER gap, nor does it fit the claim from the data it then uses as evidence. The MyST data-quality filter uses Whisper-largeV2, and Whisper models are later evaluated on the filtered data; this is a potential selection-bias confound for Whisper comparisons, but the filter is a fixed preprocessing rule applied before training and is not used to fit any parameter or to define any predicted outcome. The fine-tuning baselines (OWSM, Whisper) and zero-shot scaling results are evaluated directly on held-out test sets rather than being imported by citation as proof. Self-citations appear, but they identify tools and prior models, and none is load-bearing in the sense of replacing an independent derivation. No equation or construction maps an output back onto its own input, so the appropriate finding is no significant circularity.

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

The benchmark introduces no new theoretical entities. The free parameters are experimental design choices (upsample ratios, cluster count, CTC weight) that affect the reported numbers but do not directly encode the central claims. The axioms are the empirical assumptions needed to interpret the measurements as evidence for adult SSL bias and a scaling plateau.

free parameters (3)
  • upsample ratios = 4x for age 4-7, 2x for age 11-15
    Chosen by hand to balance speaker distribution in multi-corpus and age-group experiments (Section 4.4, Tables 4-5). The choice is not swept or justified against alternatives and affects the reported improvements.
  • K-means cluster count = 2000
    Used for discrete WavLM tokenization in Section 4.3. The paper notes that children's speech may need different granularity but does not sweep this value, so the discrete versus continuous comparison depends on this arbitrary choice.
  • CTC weight = 0.3
    Used for hybrid CTC/AED and CTC/RNNT training in Section 2.1. This is a standard value but is not varied, and it affects all flat-start model results.
assumptions (3)
  • domain assumption Whisper-largeV2 filtering of MyST does not systematically remove the hardest child utterances
    Section 3.1 filters MyST by keeping only utterances with Whisper-largeV2 WER <= 50%, duration 0.5-30s, and at least 3 words. If this filter removes a non-representative subset, the benchmark is biased toward utterances that a proprietary adult-biased model finds easy.
  • domain assumption Frozen SSL features are a valid probe of representation bias
    The adult-bias conclusion (Section 4.2) is drawn from flat-start models that use fixed, frozen WavLM and XEUS features. If those features were fine-tuned, adapters were added, or the decoder were scaled to match, the relative ranking could change. Without testing these alternatives, attributing the gap to pre-training bias is an unverified assumption.
  • domain assumption Zero-shot WER is a representative proxy for model scaling on child speech under fine-tuning
    Section 4.5 evaluates scaling only in a zero-shot setting and then concludes 'model scaling benefits child ASR up to 1B parameters'. The paper does not show that fine-tuning large models follows the same trend, so the scaling plateau claim is restricted in scope but stated generally.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Benchmarking Training Paradigms, Dataset Composition, and Model Scaling for Child ASR in ESPnet." pith.science (2026). https://pith.science/paper/QXU6MSR5

@misc{pith2026250816576,
  author       = {Pith},
  title        = {Pith review of: Benchmarking Training Paradigms, Dataset Composition, and Model Scaling for Child ASR in ESPnet},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QXU6MSR5}},
  note         = {Machine review of arXiv:2508.16576}
}
read the original abstract

Despite advancements in ASR, child speech recognition remains challenging due to acoustic variability and limited annotated data. While fine-tuning adult ASR models on child speech is common, comparisons with flat-start training remain underexplored. We compare flat-start training across multiple datasets, SSL representations (WavLM, XEUS), and decoder architectures. Our results show that SSL representations are biased toward adult speech, with flat-start training on child speech mitigating these biases. We also analyze model scaling, finding consistent improvements up to 1B parameters, beyond which performance plateaus. Additionally, age-related ASR and speaker verification analysis highlights the limitations of proprietary models like Whisper, emphasizing the need for open-data models for reliable child speech research. All investigations are conducted using ESPnet, and our publicly available benchmark provides insights into training strategies for robust child speech processing.

Figures

Figures reproduced from arXiv: 2508.16576 by the authors.

Figure 1
Figure 1. Zero-shot WER Scaling of OWLS and Whisper Models and models. Flat-start Fbank models show notable improve￾ments (rows 2-1 to 2-4 vs. 4-1 to 4-4), while fine-tuned models also benefit from broader training data (rows 2-7, 2-8 vs. 4-7, 4-8). However, for flat-start models using learned representa￾tions (rows 2-5, 2-6 vs. 4-5, 4-6), performance trends vary by dataset showing significant gains in some cases but slight d… view at source ↗

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. CHILDES-Aligned: A Curated Children's Speech Dataset via Multi-Model Timestamp Ensembling

    eess.AS 2026-07 conditional novelty 6.0 of 10

    Multi-model ASR consensus (BEACON) curates 413 h of CHILDES with corrected timestamps; the 283 h ASR subset yields up to 19.5% relative WER reduction on four held-out child benchmarks.

Reference graph

Works this paper leans on

48 extracted references · 45 canonical work pages · cited by 1 Pith paper

  1. [1]

    Introduction Large-scale speech foundation models have significantly ad- vanced automatic speech recognition (ASR), enabling sys- tems to generalize across diverse speech domains with mini- mal adaptation [1, 2, 3, 4, 5, 6]. While these advancements have yielded impressive results for adult ASR, child ASR re- mains an ongoing challenge due to the acoustic...

    work page Pith review arXiv 2025

  2. [2]

    fine-tuning), front-end representations, representation types (continuous vs

    Methodology We conduct controlled experiments to investigate key factors affecting child ASR performance: training strategies (flat-start vs. fine-tuning), front-end representations, representation types (continuous vs. discrete), dataset composition (single- vs. multi-corpus), model scaling, and age-based performance of ASR and speaker verification syste...

    work page

  3. [3]

    Child speech corpora The experiments are conducted on three child speech corpora summarized in Table 1

    Experimental setup 3.1. Child speech corpora The experiments are conducted on three child speech corpora summarized in Table 1. The MyST corpus [39] comprises dialogues between ele- mentary school students and virtual tutors across 8 science top- ics. Transcriptions include verbatim orthographic annotations, capturing hesitations, repetitions, and disflue...

    work page

  4. [4]

    Flat-start training vs

    Results and discussion 4.1. Flat-start training vs. fine-tuning Table 2 presents a comparison of flat-start training and fine- tuning, where each model is trained and evaluated on a sin- gle dataset. Among flat-start models with Fbank inputs, RNNT achieves the best performance on MyST and OGI Spon, while CTC excels on OGI Script, likely due to the limited...

    work page

  5. [5]

    While fine-tuned models generally perform best, flat-start models help mitigate biases in SSL representations, which are predominantly trained on adult speech

    Conclusion Our study comprehensively analyzes child speech processing, examining training strategies, model architectures, front-end representations, and scaling effects. While fine-tuned models generally perform best, flat-start models help mitigate biases in SSL representations, which are predominantly trained on adult speech. Multi-corpus training impr...

    work page

  6. [6]

    Additional support was provided by FWO-SBO grant S004923N: NEFL, KU Leuven C24M/22/025, and FWO grant V401325N

    Acknowledgement Experiments of this work used the Bridges2 system at PSC and Delta system at NCSA through allocations CIS210014 and IRI120008P from the Advanced Cyberinfrastructure Coordi- nation Ecosystem: Services & Support (ACCESS) program, supported by National Science Foundation grants #2138259, #tel:2138286, #tel:2138307, #tel:2137603, #tel:2138296 ...

    work page

  7. [7]

    Wavlm: Large-scale self-supervised pre-training for full stack speech processing,

    S. Chen et al., “Wavlm: Large-scale self-supervised pre-training for full stack speech processing,”IEEE Journal of Selected Topics in Signal Processing, vol. 16, no. 6, pp. 1505–1518, 2022

    2022

  8. [8]

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

    W.-N. Hsu et al., “Hubert: Self-supervised speech representation learning by masked prediction of hidden units,”IEEE/ACM Trans- actions on Audio, Speech, and Language Processing, vol. 29, pp. 3451–3460, 2021

    work page 2021

Show all 48 references
  1. [9]

    Wav2vec 2.0: A framework for self-supervised learning of speech representations,

    A. Baevski et al., “Wav2vec 2.0: A framework for self-supervised learning of speech representations,” Advances in Neural Informa- tion Processing Systems, vol. 33, pp. 12 449–12 460, 2020

    2020

  2. [10]

    Robust speech recognition via large-scale weak supervision,

    A. Radford et al., “Robust speech recognition via large-scale weak supervision,” in International Conference on Machine Learning , vol. 202, 2023, pp. 28 492–28 518

    2023

  3. [11]

    Owsm v3.1: Better and faster open whisper-style speech models based on e-branchformer,

    Y . Peng et al., “Owsm v3.1: Better and faster open whisper-style speech models based on e-branchformer,” in Interspeech, 2024, pp. 352–356

    2024

  4. [12]

    Less is more: Accurate speech recognition & translation without web-scale data,

    K. C. Puvvada et al., “Less is more: Accurate speech recognition & translation without web-scale data,” in Interspeech, 2024

    2024

  5. [13]

    Acoustics of children’s speech: Developmental changes of temporal and spectral parameters,

    S. Lee et al. , “Acoustics of children’s speech: Developmental changes of temporal and spectral parameters,” The Journal of the Acoustical Society of America, vol. 105, no. 3, pp. 1455–1468, 03 1999

    1999

  6. [14]

    On the difficulties of automatic speech recognition for kindergarten-aged children,

    G. Yeung and A. Alwan, “On the difficulties of automatic speech recognition for kindergarten-aged children,” inInterspeech, 2018, pp. 1661–1665

    2018

  7. [15]

    Challenges remain in building ASR for sponta- neous preschool children speech in naturalistic educational envi- ronments,

    S. Dutta et al., “Challenges remain in building ASR for sponta- neous preschool children speech in naturalistic educational envi- ronments,” in Interspeech, Incheon, Korea, 2022, pp. 2706–2710

    2022

  8. [16]

    V ocal tract length perturbation (vtlp) improves speech recognition,

    N. Jaitly and G. E. Hinton, “V ocal tract length perturbation (vtlp) improves speech recognition,” Proc. ICML Workshop on Deep Learning for Audio, Speech and Language, vol. 117, p. 21, 2013

    2013

  9. [17]

    Prosodic adaptations to pitch perturbation in run- ning speech,

    R. Patel et al., “Prosodic adaptations to pitch perturbation in run- ning speech,” Journal of speech, language, and hearing research, 2011

    2011

  10. [18]

    V oice Conversion Based Data Aug- mentation to Improve Children’s Speech Recognition in Limited Data Scenario,

    S. Shahnawazuddin et al., “V oice Conversion Based Data Aug- mentation to Improve Children’s Speech Recognition in Limited Data Scenario,” in Interspeech, 2020, pp. 4382–4386

    2020

  11. [19]

    Improved children’s automatic speech recognition combining adapters and synthetic data augmentation,

    T. Rolland and A. Abad, “Improved children’s automatic speech recognition combining adapters and synthetic data augmentation,” in 2024 IEEE International Conference on Acoustics, Speech and Signal Processing, 2024, pp. 12 757–12 761

    2024

  12. [20]

    Transfer learning from adult to children for speech recognition: Evaluation, analysis and recommendations,

    P. G. Shivakumar and P. Georgiou, “Transfer learning from adult to children for speech recognition: Evaluation, analysis and recommendations,” Computer speech & language , vol. 63, p. 101077, 2020

    2020

  13. [21]

    Benchmarking children’s asr with supervised and self-supervised speech foundation models,

    R. Fan et al., “Benchmarking children’s asr with supervised and self-supervised speech foundation models,” in Interspeech, 2024

    2024

  14. [22]

    Kid-whisper: Towards bridging the per- formance gap in automatic speech recognition for children vs. adults,

    A. A. Attia et al. , “Kid-whisper: Towards bridging the per- formance gap in automatic speech recognition for children vs. adults,” in Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society, vol. 7, 2024, pp. 74–80

    2024

  15. [23]

    Ml-superb 2.0: Benchmarking multilingual speech models across modeling constraints, languages, and datasets,

    J. Shi et al., “Ml-superb 2.0: Benchmarking multilingual speech models across modeling constraints, languages, and datasets,” in Interspeech, 2024, pp. 1230–1234

    2024

  16. [24]

    On the evaluation of speech foundation models for spoken language understanding,

    S. Arora et al., “On the evaluation of speech foundation models for spoken language understanding,” in Findings of ACL , Aug. 2024, pp. 11 923–11 938

    2024

  17. [25]

    Speech self-supervised representations bench- marking: a case for larger probing heads,

    S. Zaiem et al., “Speech self-supervised representations bench- marking: a case for larger probing heads,” Computer Speech & Language, vol. 89, p. 101695, 2025

    2025

  18. [26]

    Analysis of self-supervised speech models on chil- dren’s speech and infant vocalizations,

    J. Li et al., “Analysis of self-supervised speech models on chil- dren’s speech and infant vocalizations,” in 2024 IEEE Interna- tional Conference on Acoustics, Speech, and Signal Processing Workshops, 2024, pp. 550–554

    2024

  19. [27]

    Towards better domain adaptation for self- supervised models: A case study of child asr,

    R. Fan et al. , “Towards better domain adaptation for self- supervised models: A case study of child asr,” IEEE Journal of Selected Topics in Signal Processing , vol. 16, no. 6, pp. 1242– 1252, 2022

    2022

  20. [28]

    Towards universal speech discrete tokens: A case study for ASR and TTS,

    Y . Yang et al., “Towards universal speech discrete tokens: A case study for ASR and TTS,” in 2024 IEEE International Conference on Acoustics, Speech and Signal Processing , 2024, pp. 10 401– 10 405

    2024

  21. [29]

    Exploration of efficient end-to-end ASR using discretized input from self-supervised learning,

    X. Chang et al., “Exploration of efficient end-to-end ASR using discretized input from self-supervised learning,” in Interspeech. ISCA, 2023, pp. 1399–1403

    2023

  22. [30]

    A wav2vec2-based experimental study on self- supervised learning methods to improve child speech recogni- tion,

    R. Jain et al. , “A wav2vec2-based experimental study on self- supervised learning methods to improve child speech recogni- tion,” IEEE Access, vol. 11, pp. 46 938–46 948, 2023

    2023

  23. [31]

    Children’s speaker verification in low and zero resource conditions,

    S. Shahnawazuddin et al., “Children’s speaker verification in low and zero resource conditions,” Digital Signal Processing , vol. 116, p. 103115, 2021

    2021

  24. [32]

    Childaugment: Data augmentation methods for zero-resource children’s speaker verification,

    V . P. Singhet al., “Childaugment: Data augmentation methods for zero-resource children’s speaker verification,” The Journal of the Acoustical Society of America , vol. 155, no. 3, pp. 2221–2232, 2024

    2024

  25. [33]

    Effective preservation of higher-frequency contents in the context of short utterance based children’s speaker verification system,

    S. Aziz and S. Shahnawazuddin, “Effective preservation of higher-frequency contents in the context of short utterance based children’s speaker verification system,” Applied Acoustics , vol. 209, p. 109420, 2023

    2023

  26. [34]

    Owls: Scaling laws for multilingual speech recognition and translation models,

    W. Chen, J. Tian, Y . Peng, B. Yan, C.-H. H. Yang, and S. Watanabe, “Owls: Scaling laws for multilingual speech recognition and translation models,” 2025. [Online]. Available: https://arxiv.org/abs/2502.10373

    2025 arXiv

  27. [35]

    Espnet: End-to-end speech processing toolkit,

    S. Watanabe et al. , “Espnet: End-to-end speech processing toolkit,” in Interspeech, 2018, pp. 2207–2211

    2018

  28. [36]

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

    Y . Peng et al., “Reproducing whisper-style training using an open- source toolkit and publicly available data,” in ASRU, 2023, pp. 1–8

    2023

  29. [37]

    Towards robust speech representation learning for thousands of languages,

    W. Chen et al., “Towards robust speech representation learning for thousands of languages,” in Proceedings of the 2024 Con- ference on Empirical Methods in Natural Language Processing , Nov. 2024

    2024

  30. [38]

    E-branchformer: Branchformer with enhanced merging for speech recognition,

    K. Kim et al. , “E-branchformer: Branchformer with enhanced merging for speech recognition,” in SLT, 2023, pp. 84–91

    2023

  31. [39]

    A comparative study on transformer vs rnn in speech applications,

    S. Karita et al., “A comparative study on transformer vs rnn in speech applications,” in ASRU, 2019, pp. 449–456

    2019

  32. [40]

    Sequence transduction with recurrent neural net- works,

    A. Graves, “Sequence transduction with recurrent neural net- works,” ArXiv, vol. abs/1211.3711, 2012

    2012 arXiv

  33. [41]

    Connectionist temporal classification: labelling unsegmented sequence data with recurrent neural networks,

    A. Graves et al., “Connectionist temporal classification: labelling unsegmented sequence data with recurrent neural networks,” in Proceedings of the 23rd International Conference on Machine Learning, 2006, p. 369–376

    2006

  34. [42]

    OWSM-CTC: An open encoder-only speech foun- dation model for speech recognition, translation, and language identification,

    Y . Peng et al., “OWSM-CTC: An open encoder-only speech foun- dation model for speech recognition, translation, and language identification,” in Proc. ACL (Long Papers), Aug. 2024

    2024

  35. [43]

    Pushing the limits of raw waveform speaker recognition,

    J. weon Jung et al., “Pushing the limits of raw waveform speaker recognition,” in Interspeech, 2022, pp. 2228–2232

    2022

  36. [44]

    Espnet-spk: full pipeline speaker embedding toolkit with reproducible recipes, self-supervised front-ends, and off-the-shelf models,

    ——, “Espnet-spk: full pipeline speaker embedding toolkit with reproducible recipes, self-supervised front-ends, and off-the-shelf models,” in Interspeech, 2024, pp. 4278–4282

    2024

  37. [45]

    My science tutor (MyST)–a large corpus of children‘s conversational speech,

    S. Pradhan et al., “My science tutor (MyST)–a large corpus of children‘s conversational speech,” in Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation , May 2024, pp. 12 040– 12 045

    2024

  38. [46]

    The ogi kids’ speech corpus and recognizers,

    K. Shobaki et al., “The ogi kids’ speech corpus and recognizers,” Proc. of ICSLP, pp. 564–567, 2000

    2000

  39. [47]

    The cmu kids corpus,

    M. Eskenazi et al., “The cmu kids corpus,” Linguistic Data Con- sortium, vol. 11, 1997

    1997

  40. [48]

    Superb: Speech processing universal perfor- mance benchmark,

    S. wen Yang et al., “Superb: Speech processing universal perfor- mance benchmark,” in Interspeech, 2021, pp. 1194–1198

    2021

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.