Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

Employing self-supervised learning models for cross-linguistic child speech maturity classification

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read A 25-language baby-sound corpus pushes speech classification to 74.2% UAR.

desk verdict The new SpeechMaturity corpus is the real contribution and the same-test-set comparison supports the main claim, but the human-level comparison is overstated and the paper needs error bars before I'd trust the absolute numbers. read the letter →

arxiv 2506.08999 v1 pith:4FOY6KF2 submitted 2025-06-10 cs.CL cs.AI

Theo Zhang , Madurya Suresh , Anne S. Warlaumont , Kasia Hitczenko , Alejandrina Cristia , Margaret Cychosz This is my paper · ORCID
classification cs.CLcs.AI
keywords childspeechmaturityclassificationself-supervisedlearningwav2vec2cross-linguisticvocalizationcanonicalbabblingnaturalisticaudio
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 introduces SpeechMaturity, a corpus of 242,004 labeled child vocalizations from more than 25 languages and six countries, and uses it to train transformer models for a four-way classification task: cry, laughter, mature speech (consonant+vowel), and immature speech (consonant or vowel alone). The authors aim to show that a training corpus with greater linguistic and acoustic diversity produces classifiers that generalize far better than models trained on the older, smaller BabbleCorpus. Their best system reaches an unweighted average recall of 74.2% on the cleaned test set, versus 64.6% for the previous state of the art, and its agreement with human annotators is comparable to the agreement between humans themselves. If correct, this means that data scale and ecological diversity, not model sophistication alone, are what move child speech classification forward.

What carries the argument

The central object is a stack of Wav2Vec2 transformer models: a base model pre-trained on English audio, a version further pre-trained on 4,300 hours of daylong home recordings of children, and a version that adds an auxiliary child-phoneme recognition task (W2V2-LL4300-Pro). The carrying mechanism is fine-tuning these models on the SpeechMaturity corpus's 64,636-clip training set with child-disjunct folds. The auxiliary phonetic task, combined with the large diverse fine-tuning set, is what drives the classification improvement.

What would settle it

Take a random sample of about 1,000 clips from SpeechMaturity-Cleaned, have expert phoneticians (not citizen scientists) label them, and compute UAR of W2V2-LL4300-Pro-SM against the expert labels and against the citizen labels. If the model agrees with experts substantially less than with citizens, or if expert labels rearrange the class distribution, the claim that the model matches human performance would fail.

Watch

Extended reading notes

Core claim

The central discovery is that fine-tuning a Wav2Vec2 model on the diverse, naturalistic SpeechMaturity corpus yields large performance gains over the same architectures fine-tuned on BabbleCorpus, across all test sets. When the test set is held constant (the original BabbleCorpus test), SpeechMaturity fine-tuning gives 68.6% UAR versus 60.8% for BabbleCorpus fine-tuning; on the new SpeechMaturity-Cleaned test set, the best model reaches 74.2% UAR, exceeding all previously published results (best prior: 64.6%). The gain holds even for the most basic Wav2Vec2 model, indicating that the dataset's diversity is more impactful than model complexity. The best model also kept high UAR on the noisier SpeechMaturity-Uncleaned set (71.9%) and on rural recordings (67.8% versus 70.7% urban), and its per-category AUC values were strong.

Load-bearing premise

The citizen-scientist labels used as ground truth for training and evaluation are only in fair to moderate agreement with one another (weighted Fleiss kappa = 0.375 on the cleaned set), so if those labels are wrong in a class-dependent way, the reported accuracy and human-comparison numbers are not clean measurements of the models' true ability.

Editorial extensions

If this is right

  • The same architecture fine-tuned on SpeechMaturity beats all prior published results on the original BabbleCorpus test set (68.6% versus 64.6%).
  • The gain persists on noisier, uncleaned clips (71.9% UAR), so the approach should transfer to uncurated recordings.
  • Performance is similar in rural and urban environments (67.8% versus 70.7%), suggesting the classifier is not tuned to a single recording ecology.
  • Model-human agreement (weighted Cohen's kappa = 0.478) is in the same range as inter-human agreement (0.375), implying the model has reached a practical ceiling set by label noise.
  • Even the smallest model improves substantially when fine-tuned on SpeechMaturity, suggesting data diversity is the main lever.

Reading between the lines

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

  • If the dataset's label noise is the limiting factor, obtaining expert-annotated subsets could push measured performance higher, a test the paper does not run.
  • The same recipe (SSL pretraining plus large diverse fine-tuning) could be applied to other under-resourced child speech tasks, such as estimating a child's canonical babbling ratio for early screening.
  • The rural-urban gap, though small, hints at systematic acoustic differences (wind interference, overlapping speech) that could be modeled explicitly to close the gap.
  • The mature-immature distinction rests on an acoustically defined consonant-vowel transition, so the classifier may transfer across languages; that could be verified directly on held-out languages.
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 introduces SpeechMaturity, a cross-linguistic corpus of child vocalizations, and uses a subset of it to fine-tune three Wav2Vec2 models for a five-way classification task (cry, laugh, canonical, non-canonical, junk). The models are evaluated on BabbleCorpus and SpeechMaturity test sets using unweighted average recall (UAR). The best model, W2V2-LL4300-Pro-SM, reaches UAR=74.2% on SpeechMaturity-Cleaned, and on a held-constant BabbleCorpus test set SpeechMaturity fine-tuning improves UAR from 60.8% (BabbleCorpus fine-tuning) to 68.6%. The paper also reports model-human agreement (weighted Cohen's kappa 0.478) versus human-human agreement (weighted Fleiss kappa 0.375) and a rural-urban comparison.

Significance. The dataset is a potentially valuable resource for child speech research, with unprecedented language and recording-environment diversity, and the same-test-set comparison is a real and important result: fine-tuning on SpeechMaturity improves accuracy even when the test set is held constant, and the improvement appears across model architectures. The paper also states that code and data are openly available, which supports reproducibility. However, the human-comparable and absolute-accuracy claims are weakened by label-noise and metric-comparability issues, so the contribution is currently uneven and needs targeted revision.

major comments (4)
  1. [Section 4.1 and Table 3] The evaluation treats citizen-scientist majority labels as ground truth for computing UAR, yet the paper itself reports only fair-to-moderate inter-annotator agreement (weighted Fleiss kappa = 0.375 for SpeechMaturity-Cleaned and 0.271 for SpeechMaturity-Uncleaned). Moreover, Section 2.1 states that SpeechMaturity-Uncleaned labels are assigned by the highest number of annotator agreements, not necessarily a majority. If the label noise is class-dependent, the reported UAR values may reflect annotator biases rather than acoustic maturity. Please report UAR on a stricter high-agreement subset, estimate a noise ceiling (e.g., human majority-label accuracy on the same test clips), and quantify how label agreement relates to model confidence.
  2. [Section 4.1] The claim that model-human agreement 'approach and/or surpass' human-human agreement compares average weighted Cohen's kappa between the model and individual annotators (0.478) with multi-rater weighted Fleiss kappa (0.375). These values are not directly comparable because the aggregation over raters and the weighting schemes differ. Please compute human-human and model-human agreement with the same metric and the same weighting (for example, treat the model as an additional annotator in the Fleiss calculation, or compute Cohen's kappa between the model and the majority label), and report confidence intervals.
  3. [Section 4 and Table 3] No variance or significance testing is reported for the UAR values. The central same-test-set comparison (68.6 vs 60.8 on BabbleCorpus test) is based on 3,691 test clips and a single run, and Section 4 states that models 'significantly surpassed' previous state of the art without a significance test. Please provide bootstrap confidence intervals or run-to-run variance, and test whether the observed differences hold after accounting for child-level clustering.
  4. [Section 4.2 and Table 4] The rural-urban robustness claim is based on one model and the table reports only UAR with SD<0.01, with no information about the number of children, clips, languages, or recording devices in each environment. The 2.9-point UAR difference could be confounded by language, child age, or recording hardware. Please report per-environment sample sizes, a significance test, or both, or temper the robustness claim.
minor comments (5)
  1. [Abstract and Section 2.1] The abstract says the dataset contains 242,004 labeled vocalizations, but the experiments use a down-sampled subset of 64,636 clips; please clarify the relationship between the full corpus and the experimental subset.
  2. [Table 4] The total clip count in Table 4 (51,390) is inconsistent with the SpeechMaturity-Cleaned train/dev/test totals in Table 1 (53,089); please reconcile the numbers.
  3. [Figure 2] The ROC curves are not accompanied by the AUC values in the text or in a table; please report the AUC numbers explicitly.
  4. [Section 3] The description of W2V2-LL4300-Pro's auxiliary phonetic task is difficult to follow; a short architecture diagram or a precise equation would help.
  5. [Throughout] There are several typographical issues, including 'V ocalizations' in Section 2.1 and a missing '%' after UAR=62.9 in Section 4; a careful proofread is needed.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: the benchmark is an external empirical comparison, with only minor self-citations for dataset construction.

full rationale

The paper's central comparison is an empirical benchmark, not a derivation from fitted parameters. The headline result—W2V2-LL4300-Pro-SM reaching UAR 74.2% on SpeechMaturity-Cleaned and 68.6% on the held-out BabbleCorpus test set—is produced by standard supervised fine-tuning on a training split and evaluation on child-disjunct held-out splits. The comparison against prior state-of-the-art (Li et al. 64.6%) is made on the same BabbleCorpus test set, whose labels came from the 2019 Interspeech Challenge's ≥66% majority criterion and are external to the present fine-tuning. The authors' self-citations ([22], [23], [32]) support corpus construction and annotation methodology; they do not define the test labels or the test predictions. No equation in the paper defines the output in terms of the input labels or fits a parameter that is then renamed a prediction. The human-agreement comparison mixes weighted Cohen's kappa (model-vs-annotator) with weighted Fleiss' kappa (annotator-vs-annotator), which is a metric-comparability concern rather than circularity, and the reliance on citizen-scientist labels is an annotation-quality limitation that does not make the accuracy measurement equivalent to its inputs by construction.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The paper's claims rest on the quality of the citizen-science labels and the diarization step that generated the clips. The self-reported inter-annotator agreement is fair to moderate, which directly limits the interpretability of small UAR differences. The hyperparameters (learning rates, epochs, downsampling ratio) are reported but not justified by tuning experiments. No new theoretical entities are introduced.

free parameters (5)
  • Learning rate for W2V2-base = 3e-5
    Chosen in Section 3 for fine-tuning; not reported as tuned.
  • Learning rate for W2V2-LL4300h and Pro = 1e-5
    Chosen in Section 3, following the prior work's setup.
  • Number of fine-tuning epochs = 10
    Training ran 10 epochs; best epoch chosen on dev (Section 3).
  • Downsampling multiplier for class imbalance = approximately 3x the laughing class
    Section 2.1: all classes except laughing downsampled to about 3x the total laughing clips.
  • Custom kappa weighting scheme = not fully specified
    Section 4.1 uses custom weights for Fleiss and Cohen kappa; the exact weights are not given.
assumptions (4)
  • domain assumption LENA and Voice Type Classifier diarization correctly identify child vocalizations from daylong recordings
    Section 2.1: vocalizations were extracted via LENA or VTC; errors in diarization would propagate to labels.
  • domain assumption Citizen-scientist majority labels are valid ground truth for vocalization maturity
    Section 2.1 and 4.1: at least 3 annotators per clip; for SM-U labels have only plurality agreement; low inter-annotator kappa (0.271-0.457) makes this assumption fragile.
  • domain assumption Child-disjunct train/dev/test splits prevent speaker leakage
    Section 2.2 states splits are child-disjunct; if violated, reported UAR would be inflated.
  • domain assumption The 80/10/10 split with stratification by age and language yields representative test sets
    Section 2.2: authors ensured even representation, but no test for balance is reported.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Employing self-supervised learning models for cross-linguistic child speech maturity classification." pith.science (2026). https://pith.science/paper/4FOY6KF2

@misc{pith2026250608999,
  author       = {Pith},
  title        = {Pith review of: Employing self-supervised learning models for cross-linguistic child speech maturity classification},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4FOY6KF2}},
  note         = {Machine review of arXiv:2506.08999}
}
read the original abstract

Speech technology systems struggle with many downstream tasks for child speech due to small training corpora and the difficulties that child speech pose. We apply a novel dataset, SpeechMaturity, to state-of-the-art transformer models to address a fundamental classification task: identifying child vocalizations. Unlike previous corpora, our dataset captures maximally ecologically-valid child vocalizations across an unprecedented sample, comprising children acquiring 25+ languages in the U.S., Bolivia, Vanuatu, Papua New Guinea, Solomon Islands, and France. The dataset contains 242,004 labeled vocalizations, magnitudes larger than previous work. Models were trained to distinguish between cry, laughter, mature (consonant+vowel), and immature speech (just consonant or vowel). Models trained on the dataset outperform state-of-the-art models trained on previous datasets, achieved classification accuracy comparable to humans, and were robust across rural and urban settings.

Figures

Figures reproduced from arXiv: 2506.08999 by the authors.

Figure 1
Figure 1. Confusion matrix of the SpeechMaturity-Cleaned test set predictions for W2V2-LL4300-Pro finetuned on SpeechMaturity-Cleaned. vious work has not, we conducted an agreement analysis on W2V2-LL4300-Pro-SM. We assessed inter-human annota￾tor agreement using a weighted Fleiss’ kappa metric for the SpeechMaturity-Cleaned and -Uncleaned datasets. We em￾ployed a custom weighting scheme to reflect the relative im￾portance of… view at source ↗
Figure 2
Figure 2. ROC curves achieved by W2V2-LL4300-Pro-SM on SpeechMaturity-Cleaned (top) and W2V2-LL4300-Pro-SM on SpeechMaturity-Uncleaned (bottom) for classification of canonical (green), non-canonical (red), crying (yellow), laugh￾ing (pink), and junk (blue) in comparison to human annotators. Next, we compared human annotator performance to W2V2-LL4300-Pro-SM by calculating weighted Cohen’s kappa between W2V2-LL4300-Pro-SM and … view at source ↗

Discussion (0). Continue with ORCID 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. Deriving Benchmarking Datasets from Long-Form Recordings: Challenges and Opportunities

    eess.AS 2026-07 accept novelty 6.0 of 10

    A mutually dependent framework standardizes 27 LFR child-speech corpora, builds four benchmarks, and embeds ELSI governance, with a VTC case study showing private data is needed for competitive performance.

Reference graph

Works this paper leans on

37 extracted references · 36 canonical work pages · cited by 1 Pith paper

  1. [1]

    By about 6-7 months of age, infants start producing sounds that contain both consonant and vowel elements, form- ing what are known as canonical syllables [1]

    Introduction and related work In the first years of life, children’s speech becomes increasingly adult-like. By about 6-7 months of age, infants start producing sounds that contain both consonant and vowel elements, form- ing what are known as canonical syllables [1]. Canonical sylla- bles continue to make up an increasing proportion of children’s vocal p...

    work page 2019

  2. [2]

    Employing self-supervised learning models for cross-linguistic child speech maturity classification

    Child speech corpora 2.1. Corpora construction We employ two different training corpora: BabbleCor- pus (N=11,304 labeled vocalizations from 6 languages)— the training dataset for the current baseline models—and the significantly-expanded SpeechMaturity (N=64,636 vocaliza- tions from 25+ languages) which has never been employed for this task before. Babbl...

    work page Pith review arXiv 2025

  3. [3]

    Model architectures We employed three different Wav2Vec2 models of varying size and complexity for the task of child speech maturity classifi- cation:W2V2-base,W2V2-LL4300h, andW2V2-LL4300-Pro. The three models were pretrained in different ways and fine- tuned on either the BabbleCorpus or SpeechMaturity-Cleaned datasets (explained below).W2V2-basewas pre...

    work page

  4. [4]

    canonical

    Results Following previous classification models for this task (e.g. [26]), model performance was evaluated using the unweighted average recall (UAR), a metric that takes the mean of recall val- ues for each class, giving equal importance to each class re- gardless of its size. It is thus well-suited for multi-class clas- sification tasks, especially when...

    work page

  5. [5]

    Discussion SpeechMaturity represents a significant shift in child speech re- search, challenging existing methodological constraints in com- putational studies of child speech development. By capturing child vocalizations across 25+ languages and dramatically di- verse acoustic environments—from urban centers in industrial- ized communities to remote comm...

    work page

  6. [6]

    D. K. Oller,The Emergence of the Speech Capacity. Mahwah, NJ: Lawrence Erlbaum Associates, 2000

    work page 2000

  7. [7]

    Automatic Detec- tion of Speech Sound Disorder in Child Speech Using Posterior- based Speaker Representations,

    S.-I. Ng, C. W.-Y . Ng, J. Wang, and T. Lee, “Automatic Detec- tion of Speech Sound Disorder in Child Speech Using Posterior- based Speaker Representations,” inInterspeech 2022. ISCA, Sep. 2022, pp. 2853–2857

    2022

  8. [8]

    Maddieson,Patterns of Sounds, ser

    I. Maddieson,Patterns of Sounds, ser. Cambridge Studies in Speech Science and Communication. Cambridge [Cam- bridgeshire] ; New York: Cambridge University Press, 1984

    work page 1984

Show all 37 references
  1. [9]

    Adaptation to Language: Evidence from Babbling and First Words in Four Languages,

    B. de Boysson-Bardies, M. M. Vihman, and B. de Boysson- Bardies, “Adaptation to Language: Evidence from Babbling and First Words in Four Languages,”Language, vol. 67, no. 2, pp. 297–319, Jun. 1991

    1991

  2. [10]

    Towards Better Do- main Adaptation for Self-Supervised Models: A Case Study of Child ASR,

    R. Fan, Y . Zhu, J. Wang, and A. Alwan, “Towards Better Do- main Adaptation for Self-Supervised Models: A Case Study of Child ASR,”IEEE Journal of Selected Topics in Signal Process- ing, vol. 16, no. 6, pp. 1242–1252, Oct. 2022

    2022

  3. [11]

    Self-supervised learning for infant cry analysis,

    A. Gorin, C. Subakan, S. Abdoli, J. Wang, S. Latremouille, and C. Onu, “Self-supervised learning for infant cry analysis,” in2023 IEEE International Conference on Acoustics, Speech, and Signal Processing Workshops (ICASSPW), 2023, pp. 1–5

    2023

  4. [12]

    Enhancing Child V ocalization Classification with Phonetically-Tuned Em- beddings for Assisting Autism Diagnosis,

    J. Li, M. Hasegawa-Johnson, and K. Karahalios, “Enhancing Child V ocalization Classification with Phonetically-Tuned Em- beddings for Assisting Autism Diagnosis,” Jun. 2024

    2024

  5. [13]

    VCMNet: Weakly Supervised Learning for Auto- matic Infant V ocalisation Maturity Analysis,

    N. Al Futaisi, Z. Zhang, A. Cristia, A. Warlaumont, and B. Schuller, “VCMNet: Weakly Supervised Learning for Auto- matic Infant V ocalisation Maturity Analysis,” in2019 Interna- tional Conference on Multimodal Interaction. Suzhou China: ACM, Oct. 2019, pp. 205–209

    2019

  6. [14]

    Automated Classification of Children’s Linguistic versus Non- Linguistic V ocalisations,

    Z. Zhang, A. Cristia, A. S. Warlaumont, and B. Schuller, “Automated Classification of Children’s Linguistic versus Non- Linguistic V ocalisations,” inProceedings of Interspeech 2018, Hyderabad, India, 2018

    2018

  7. [15]

    The INTERSPEECH 2019 Computational Par- alinguistics Challenge: Styrian Dialects, Continuous Sleepiness, Baby Sounds & Orca Activity,

    B. W. Schuller, A. Batliner, C. Bergler, F. B. Pokorny, J. Krajewski, M. Cychosz, R. V ollmann, S.-D. Roelen, S. Schnieder, E. Bergelson, A. Cristia, A. Seidl, A. S. War- laumont, L. Yankowitz, E. N ¨oth, S. Amiriparian, S. Hantke, and M. Schmitt, “The INTERSPEECH 2019 Computa...

    2019

  8. [16]

    Vali- dating a model to detect infant crying from naturalistic audio,

    M. Micheletti, X. Yao, M. Johnson, and K. De Barbaro, “Vali- dating a model to detect infant crying from naturalistic audio,” Behavior Research Methods, vol. 55, no. 6, pp. 3187–97, 2022

    2022

  9. [17]

    Towards Ro- bust Family-Infant Audio Analysis Based on Unsupervised Pre- training of Wav2vec 2.0 on Large-Scale Unlabeled Family Au- dio,

    J. Li, M. Hasegawa-Johnson, and N. L. McElwain, “Towards Ro- bust Family-Infant Audio Analysis Based on Unsupervised Pre- training of Wav2vec 2.0 on Large-Scale Unlabeled Family Au- dio,” inINTERSPEECH 2023, Aug. 2023, pp. 1035–1039

    2023

  10. [18]

    Casillas, P

    M. Casillas, P. Brown, and S. Levinson,Casillas HomeBank Cor- pus, 2017

    2017

  11. [19]

    Cychosz,Cychosz HomeBank Corpus, 2018

    M. Cychosz,Cychosz HomeBank Corpus, 2018

    2018

  12. [20]

    Bergelson,Bergelson Seedlings HomeBank Corpus, 2017

    E. Bergelson,Bergelson Seedlings HomeBank Corpus, 2017

    2017

  13. [21]

    Cristia and H

    A. Cristia and H. Colleran,Long-Form, Child-Centered Record- ings Collected in Malekula in 2016-2018, 2018

    2016

  14. [22]

    Warlaumont, G

    A. Warlaumont, G. Pretzer, S. Mendoza, and E. Walle,Warlau- mont HomeBank Corpus, 2016

    2016

  15. [23]

    Scaff, J

    C. Scaff, J. Stieglitz, and A. Cristi `a,Daylong Recordings from Young Children Learning Tsimane in Bolivia, 2018

    2018

  16. [24]

    HomeBank, an online repository of daylong child-centered audio recordings,

    M. VanDam, A. S. Warlaumont, E. Bergelson, A. Cristi `a, M. Soderstrom, P. De Palma, and B. MacWhinney, “HomeBank, an online repository of daylong child-centered audio recordings,” Seminars in Speech and Language, vol. 37, pp. 128–142, 2016

    2016

  17. [25]

    Reliability of the LENA Lan- guage Environment Analysis System in young children’s natural home environment,

    D. Xu, U. Yapanel, and S. Gray, “Reliability of the LENA Lan- guage Environment Analysis System in young children’s natural home environment,” LENA Research Foundation, Boulder, CO, Technical Report lTR-05-2, 2009

    2009

  18. [26]

    An open-source voice type classifier for child-centered daylong recordings,

    M. Lavechin, R. Bousbib, H. Bredin, E. Dupoux, and A. Cristia, “An open-source voice type classifier for child-centered daylong recordings,”arXiv:2005.12656 [eess], 2021

    2005 arXiv

  19. [27]

    V o- cal development in a large-scale crosslinguistic corpus,

    M. Cychosz, A. Cristia, E. Bergelson, M. Casillas, G. Baudet, A. S. Warlaumont, C. Scaff, L. Yankowitz, and A. Seidl, “V o- cal development in a large-scale crosslinguistic corpus,”Develop- mental Science, vol. 24, no. 5, p. e13090, 2021

    2021

  20. [28]

    The development of canonical proportion continues past toddlerhood,

    K. Hitczenko, E. Bergelson, M. Casillas, H. Colleran, M. Cy- chosz, and A. Cristia, “The development of canonical proportion continues past toddlerhood,” inProceedings of the International Congress of the Phonetic Sciences, Prague, CZ, 2023

    2023

  21. [29]

    Lib- rispeech: An ASR corpus based on public domain audio books

    V . Panayotov, G. Chen, D. Povey, and S. Khudanpur, “Lib- rispeech: An ASR corpus based on public domain audio books.”

  22. [30]

    Wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Represen- tations,

    A. Baevski, H. Zhou, A. Mohamed, and M. Auli, “Wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Represen- tations,” inProceedings of the 34th International Conference on NeurIPS Systems, 2020, pp. 12 449–12 460

    2020

  23. [31]

    Using Attention Networks and Adversar- ial Augmentation for Styrian Dialect Continuous Sleepiness and Baby Sound Recognition,

    S. L. Yeh, G.-Y . Chao, B. Su, Y .-L. Huang, M.-H. Lin, Y .-C. Tsai, Y .-W. Tai, Z.-C. Lu, C.-Y . Chen, T.-M. Tai, C.-W. Tseng, C.-K. Lee, and C.-C. Lee, “Using Attention Networks and Adversar- ial Augmentation for Styrian Dialect Continuous Sleepiness and Baby Sound Recogniti...

    2019

  24. [32]

    Keesing, Y

    A. Keesing, Y . Koh, and M. Witbrock,Acoustic Features and Neu- ral Representations for Categorical Emotion Recognition from Speech, Aug. 2021

    2021

  25. [33]

    Using Fisher Vector and Bag-of-Audio-Words Representations to Identify Styrian Dialects, Sleepiness, Baby & Orca Sounds,

    G. Gosztolya, “Using Fisher Vector and Bag-of-Audio-Words Representations to Identify Styrian Dialects, Sleepiness, Baby & Orca Sounds,” inInterspeech 2019. ISCA, Sep. 2019, pp. 2413– 2417

    2019

  26. [34]

    Com- bining Clustering and Functionals based Acoustic Feature Rep- resentations for Classification of Baby Sounds,

    H. Kaya, O. Verkholyak, M. Markitantov, and A. Karpov, “Com- bining Clustering and Functionals based Acoustic Feature Rep- resentations for Classification of Baby Sounds,” inCompanion Publication of the 2020 International Conference on Multimodal Interaction. Virtual Event Net...

    2020

  27. [35]

    Measuring nominal scale agreement among many raters,

    J. L. Fleiss, “Measuring nominal scale agreement among many raters,”Psychological Bulletin, vol. 76, no. 5, pp. 378–382, 1971

    1971

  28. [36]

    Characteriza- tion of children’s verbal input in a forager-farmer population us- ing long-form audio recordings and diverse input definitions,

    C. Scaff, M. Casillas, J. Stieglitz, and A. Cristia, “Characteriza- tion of children’s verbal input in a forager-farmer population us- ing long-form audio recordings and diverse input definitions,”In- fancy, vol. n/a, no. n/a

  29. [37]

    Speech Maturity Dataset: A cross- cultural corpus of naturalistic child and adult vocalizations,

    K. Hitczenko, L. Peurey, W. Havard, Tey, A. Seidl, C. Semenzin, M. Lavechin, B. Kelleher, L. Hamrick, L. Gautheron, M. Cychosz, M. Casillas, and A. Cristia, “Speech Maturity Dataset: A cross- cultural corpus of naturalistic child and adult vocalizations,” under review

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.