pith. sign in

arxiv: 2606.01134 · v1 · pith:TG3P34B4new · submitted 2026-05-31 · 📡 eess.AS · cs.LG· cs.SD

Context-aware child-directed speech detection from long-form recordings

Th\'eo Charlot , Tarek Kunze , Kaveri K. Sheth , Alejandrina Cristia , Marvin Lavechin This is my paper

Pith reviewed 2026-06-28 16:39 UTC · model grok-4.3

classification 📡 eess.AS cs.LGcs.SD
keywords child-directed speechadult-directed speechlong-form recordingscontext-aware classificationself-supervised speech modelsmultilingual child dataaddressee detectionspeech segmentation
0
0 comments X

The pith

Incorporating surrounding context and in-domain pre-training substantially improves detection of child-directed speech from long-form recordings.

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

The paper shows that adding audio context around each utterance and pre-training self-supervised models on child-centered data markedly raises accuracy when separating speech directed at children from adult-directed speech. This matters for automatic, large-scale measurement of children's language input without exhaustive manual labeling of every recording. Tests on a multilingual set of 182 children demonstrate consistent gains over isolated-utterance processing and over models trained only on adult speech. The same models still outperform a rule-based baseline when embedded in a full pipeline that starts with automatic adult-speech detection.

Core claim

Fine-tuning six self-supervised models on child-centered recordings and feeding surrounding context into the classifier produces an absolute 13.8 percent gain in average F1-score for child-directed versus adult-directed speech classification, outperforming both context-free baselines and adult-speech pre-trained models, with usable though reduced performance retained in an end-to-end pipeline from speech detection onward.

What carries the argument

The incorporation of surrounding audio context into the classification step, applied to models that have first undergone in-domain pre-training on child-centered recordings.

If this is right

  • In-domain pre-training on child-centered recordings yields higher accuracy than pre-training on adult speech alone.
  • Context from neighboring utterances raises average F1 by 13.8 percentage points.
  • An end-to-end pipeline that includes automatic segmentation still beats a rule-based baseline even after segmentation errors.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The method could support larger cross-linguistic studies of children's everyday language exposure without proportional increases in manual annotation effort.
  • Performance may degrade further if the target recordings contain heavier background noise or different microphone placements than the training set.
  • Extending the context window size or combining it with speaker diarization outputs could produce additional gains.

Load-bearing premise

The manual labels supplied with the 182-child multilingual dataset correctly identify which utterances are directed at the child.

What would settle it

Running the context-aware model on a fresh collection of long-form recordings whose addressee labels were produced by multiple independent human listeners and observing no F1 improvement relative to a context-free model.

Figures

Figures reproduced from arXiv: 2606.01134 by Alejandrina Cristia, Kaveri K. Sheth, Marvin Lavechin, Tarek Kunze, Th\'eo Charlot.

Figure 1
Figure 1. Figure 1: Effect of context duration on the validation F1-score (%), where 0 seconds corresponds to no context, i.e., the model receives only the target utterance. Standard deviations are com￾puted across 5 random seeds. stable and not an artifact of hyperparameter selection. We re￾fer to this model as BabyHuBERT-addressee in the remaining analysis. 3.3. Assessing performance in realistic conditions So far, our anal… view at source ↗
read the original abstract

Automatically distinguishing child-directed speech from adult-directed speech in long-form recordings is key to scalable analyses of children's language environments. Existing approaches process utterances in isolation and have been evaluated primarily on English. We address these gaps along three dimensions. First, we fine-tune and evaluate six-self supervised models on a multilingual dataset of 182 children, showing that in-domain pre-training on child-centered recordings substantially outperforms models trained on adult speech. Second, we demonstrate that incorporating surrounding context substantially improves classification, with an absolute gain of 13.8% in average F1-score. Third, we evaluate our model in a realistic end-to-end pipeline, from adult speech detection to addressee classification, showing that performance drops under automatic segmentation but still consistently outperforms a rule-based baseline.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The paper evaluates six self-supervised speech models fine-tuned on a multilingual dataset of 182 children for distinguishing child-directed from adult-directed speech in long-form recordings. It reports that in-domain pre-training outperforms adult-speech models, that adding surrounding context yields a 13.8% absolute gain in average F1-score, and that the resulting system outperforms a rule-based baseline in a realistic end-to-end pipeline from adult-speech detection to addressee classification.

Significance. If the empirical gains hold under proper validation, the work would provide a practical advance for scalable, automated analysis of children's language environments, extending prior isolated-utterance approaches to multilingual long-form data and demonstrating the value of context and domain-matched pre-training.

major comments (3)
  1. [Abstract and results] Abstract (paragraph 3) and results section: the headline 13.8% absolute F1 gain from context and the superiority of in-domain pre-training are reported without error bars, statistical tests, or any description of how context is encoded (e.g., concatenation window, attention mechanism) or how train/test splits were performed across the 182 children. These omissions make the central performance claims impossible to assess for robustness.
  2. [Dataset description] Dataset description (abstract paragraph 2 and methods): the multilingual corpus supplies the sole ground truth for both training and evaluation, yet no inter-annotator agreement statistics, label-distribution tables, or error analysis for ambiguous cases (overlapping speech, cultural prosody variation) are provided. Because all reported F1 numbers rest on these manual labels, their reliability is load-bearing.
  3. [Pipeline evaluation] End-to-end pipeline evaluation: the text states that performance drops under automatic segmentation but remains above the rule-based baseline; however, no quantitative breakdown of the segmentation error contribution versus the classification error is given, preventing evaluation of whether the context-aware component actually drives the reported improvement in the realistic setting.
minor comments (2)
  1. [Methods] Model names and pre-training corpora should be listed explicitly in a table rather than referenced only by citation.
  2. [Figures] Figure captions for the pipeline diagram should clarify the exact input/output of each stage.

Simulated Author's Rebuttal

3 responses · 1 unresolved

We thank the referee for their constructive feedback. We address each major comment below, indicating where we will revise the manuscript to improve clarity and robustness.

read point-by-point responses

  1. Referee: [Abstract and results] Abstract (paragraph 3) and results section: the headline 13.8% absolute F1 gain from context and the superiority of in-domain pre-training are reported without error bars, statistical tests, or any description of how context is encoded (e.g., concatenation window, attention mechanism) or how train/test splits were performed across the 182 children. These omissions make the central performance claims impossible to assess for robustness.

    Authors: We agree that additional details would strengthen the claims. The train/test splits are speaker-independent across the 182 children (detailed in Methods). Context is incorporated via concatenation of a fixed window of neighboring utterances, processed through the model's attention layers. In revision we will add error bars, statistical tests (e.g., paired significance tests across folds), and an expanded description of the context-encoding procedure. revision: yes

  2. Referee: [Dataset description] Dataset description (abstract paragraph 2 and methods): the multilingual corpus supplies the sole ground truth for both training and evaluation, yet no inter-annotator agreement statistics, label-distribution tables, or error analysis for ambiguous cases (overlapping speech, cultural prosody variation) are provided. Because all reported F1 numbers rest on these manual labels, their reliability is load-bearing.

    Authors: Labels originate from the existing corpus. We cannot supply inter-annotator agreement because it was not computed or released by the corpus providers. We will add label-distribution tables and a short discussion of ambiguous cases (e.g., overlapping speech) based on available metadata. revision: partial

  3. Referee: [Pipeline evaluation] End-to-end pipeline evaluation: the text states that performance drops under automatic segmentation but remains above the rule-based baseline; however, no quantitative breakdown of the segmentation error contribution versus the classification error is given, preventing evaluation of whether the context-aware component actually drives the reported improvement in the realistic setting.

    Authors: We will add a quantitative error breakdown in the pipeline section that isolates segmentation error from classification error, allowing readers to assess the contribution of the context-aware model under automatic segmentation. revision: yes

standing simulated objections not resolved
  • Inter-annotator agreement statistics for the corpus labels, which are unavailable from the original annotations.

Circularity Check

0 steps flagged

No circularity: purely empirical results on held-out data

full rationale

The paper reports empirical measurements from fine-tuning and evaluating self-supervised models on a held-out portion of a manually labeled multilingual dataset of 182 children. No equations, derivations, fitted parameters renamed as predictions, or self-citation chains are present in the abstract or described claims. Performance gains (e.g., 13.8% F1 from context) are presented as direct experimental outcomes rather than reductions to training inputs or prior self-authored uniqueness results. The work is self-contained against external benchmarks and does not invoke any of the enumerated circularity patterns.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review supplies no explicit free parameters, axioms, or invented entities; standard machine-learning assumptions (i.i.d. splits, label accuracy, transferability of self-supervised representations) are implicit but unstated.

pith-pipeline@v0.9.1-grok · 5676 in / 1125 out tokens · 24710 ms · 2026-06-28T16:39:16.327036+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

50 extracted references · 2 canonical work pages · 2 internal anchors

  1. [1]

    Introduction Children’s environments are complex, and the language input they receive is no exception. Among the sources of this com- plexity is the distinction between child-directed speech (CDS), the register adults typically adopt when speaking to young chil- dren, and adult-directed speech (ADS). CDS is characterized by features such as higher pitch, ...

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  2. [2]

    We then formal- ize the addressee classification problem before introducing the self-supervised models we considered, and the context-aware fine-tuning strategy we implemented

    Methods We introduce the corpora used in this study. We then formal- ize the addressee classification problem before introducing the self-supervised models we considered, and the context-aware fine-tuning strategy we implemented. We present the baseline against which our best model is compared and the evaluation metric. We conclude this section by providi...

  3. [3]

    Effect of self-supervised models We begin by addressing our first question, comparing multiple self-supervised models fine-tuned on our addressee classifica- tion task (Table 2)

    Results 3.1. Effect of self-supervised models We begin by addressing our first question, comparing multiple self-supervised models fine-tuned on our addressee classifica- tion task (Table 2). Among out-of-domain models pre-trained on adult speech, W2V2, HuBERT and W2V2-XLSR achieve comparable F1- scores in the 45% - 57% range, suggesting no clear benefit ...

  4. [4]

    Discussion and conclusion Our results show that large-scale, automatic detection ofwho speaks to the childfrom naturalistic long-form recordings is fea- sible. Importantly, our results highlight two key factors for im- proving performance: domain-matched multilingual pretrain- ing, with BabyHuBERT consistently outperforming other self- supervised models, ...

  5. [5]

    TC was funded by an ERC grant (InfantSimu- lator, 101142705); AC, KS and TK were funded by an ERC grant (ExELang, 101001095)

    Acknowledgments This work was performed using HPC resources from GENCI- IDRIS (Grant 2024-AD01101545 and 2025-AD011016414) and was supported in part by the Agence Nationale pour la Recherche (ANR-17-EURE-0017 Frontcog, ANR10-IDEX- 0001-02 PSL). TC was funded by an ERC grant (InfantSimu- lator, 101142705); AC, KS and TK were funded by an ERC grant (ExELang...

    2024

  6. [6]

    The inevitability of child directed speech,

    M. Saxton, “The inevitability of child directed speech,” inLan- guage acquisition. Springer, 2009, pp. 62–86

    2009

  7. [7]

    Acoustic-lexical characteristics of child-directed speech between 7 and 24 months and their impact on toddlers’ phonological processing,

    M. Cychosz, J. R. Edwards, N. Bernstein Ratner, C. Torring- ton Eaton, and R. S. Newman, “Acoustic-lexical characteristics of child-directed speech between 7 and 24 months and their impact on toddlers’ phonological processing,”Frontiers in Psychology, vol. 12, p. 712647, 2021

    2021

  8. [8]

    Does child-directed speech facilitate language development in all domains? a study space analysis of the existing evidence,

    V . Kempe, M. Ota, and S. Schaeffler, “Does child-directed speech facilitate language development in all domains? a study space analysis of the existing evidence,”Developmental Review, vol. 72, p. 101121, 2024

    2024

  9. [9]

    Word segmentation cues in German child-directed speech: A corpus analysis,

    K. St ¨ark, E. Kidd, and R. L. Frost, “Word segmentation cues in German child-directed speech: A corpus analysis,”Language and Speech, vol. 65, no. 1, pp. 3–27, 2022

    2022

  10. [10]

    Quantifying sources of variability in infancy re- search using the infant-directed-speech preference,

    M. Consortium, “Quantifying sources of variability in infancy re- search using the infant-directed-speech preference,”Advances in Methods and Practices in Psychological Science, vol. 3, no. 1, pp. 24–52, 2020

    2020

  11. [11]

    Statistical speech seg- mentation and word learning in parallel: Scaffolding from child- directed speech,

    D. Yurovsky, C. Yu, and L. B. Smith, “Statistical speech seg- mentation and word learning in parallel: Scaffolding from child- directed speech,”Frontiers in psychology, vol. 3, p. 374, 2012

    2012

  12. [13]

    Language learning, socioeco- nomic status, and child-directed speech,

    J. F. Schwab and C. Lew-Williams, “Language learning, socioeco- nomic status, and child-directed speech,”Wiley Interdisciplinary Reviews: Cognitive Science, vol. 7, no. 4, pp. 264–275, 2016

    2016

  13. [14]

    The INTERSPEECH 2017 Computa- tional Paralinguistics Challenge: Addressee, Cold & Snoring,

    B. Schuller, S. Steidl, A. Batliner, E. Bergelson, J. Krajewski, C. Janott, A. Amatuni, M. Casillas, A. Seidl, M. Soderstrom, A. S. Warlaumont, G. Hidalgo, S. Schnieder, C. Heiser, W. Hohenhorst, M. Herzog, M. Schmitt, K. Qian, Y . Zhang, G. Trigeorgis, P. Tzi- rakis, and S. Zafeiriou, “The INTERSPEECH 2017 Computa- tional Paralinguistics Challenge: Addre...

    2017

  14. [15]

    DNN- Based Feature Extraction and Classifier Combination for Child- Directed Speech, Cold and Snoring Identification,

    G. Gosztolya, R. Busa-Fekete, T. Gr ´osz, and L. T ´oth, “DNN- Based Feature Extraction and Classifier Combination for Child- Directed Speech, Cold and Snoring Identification,” inInterspeech, 2017, pp. 3522–3526

    2017

  15. [16]

    Introducing Weighted Kernel Clas- sifiers for Handling Imbalanced Paralinguistic Corpora: Snoring, Addressee and Cold,

    H. Kaya and A. A. Karpov, “Introducing Weighted Kernel Clas- sifiers for Handling Imbalanced Paralinguistic Corpora: Snoring, Addressee and Cold,” inInterspeech, 2017, pp. 3527–3531

    2017

  16. [17]

    An automated classifier for child-directed speech from lena record- ings,

    J. Y . Bang, G. Kachergis, A. Weisleder, and V . A. Marchman, “An automated classifier for child-directed speech from lena record- ings,” inProceedings of the 46th annual Boston University Con- ference on Language Development, Y . Gong and F. Kpogo, Eds. Somerville, MA: Cascadilla Press, 2022, pp. 48–61

    2022

  17. [18]

    Hearttoheart: The arts of infant versus adult-directed speech classification,

    N. D. Al Futaisi, A. Cristia, and B. W. Schuller, “Hearttoheart: The arts of infant versus adult-directed speech classification,” in International Conference on Acoustics, Speech and Signal Pro- cessing, 2023, pp. 1–5

    2023

  18. [19]

    The weirdest people in the world?

    J. Henrich, S. J. Heine, and A. Norenzayan, “The weirdest people in the world?”Behavioral and Brain Sciences, vol. 33, no. 2–3, p. 61–83, 2010

    2010

  19. [20]

    Child- directed speech is infrequent in a forager-farmer population: A time allocation study,

    A. Cristia, E. Dupoux, M. Gurven, and J. Stieglitz, “Child- directed speech is infrequent in a forager-farmer population: A time allocation study,”Child Development, vol. 90, no. 3, pp. 759– 773, 2019

    2019

  20. [21]

    Babyhubert: Multilingual self-supervised learning for segmenting speakers in child-centered long-form recordings,

    T. Charlot, T. Kunze, M. Poli, A. Cristia, E. Dupoux, and M. Lavechin, “Babyhubert: Multilingual self-supervised learning for segmenting speakers in child-centered long-form recordings,”

  21. [22]

    BabyHuBERT: Multilingual Self-Supervised Learning for Segmenting Speakers in Child-Centered Long-Form Recordings

    [Online]. Available: https://arxiv.org/abs/2509.15001

    work page internal anchor Pith review arXiv

  22. [23]

    Homebank: An online repository of daylong child-centered audio recordings,

    M. VanDam, A. S. Warlaumont, E. Bergelson, A. Cristia, M. Soderstrom, P. De Palma, and B. MacWhinney, “Homebank: An online repository of daylong child-centered audio recordings,” Semin Speech Lang, vol. 37, no. 02, pp. 128–142, 2016

    2016

  23. [24]

    Ticuna (tca) language documentation: A guide to ma- terials in the california language archive,

    A. Skilton, “Ticuna (tca) language documentation: A guide to ma- terials in the california language archive,”Language Documenta- tion and Conservation, vol. 15, pp. 153–189, 2021

    2021

  24. [25]

    MacWhinney,The CHILDES project, 3rd ed

    B. MacWhinney,The CHILDES project, 3rd ed. London, Eng- land: Psychology Press, 2014

    2014

  25. [26]

    Lyon HomeBank Corpus,

    M. Canault, M.-T. Le Normand, S. Foudil, N. Loundon, and H. Thai-Van, “Lyon HomeBank Corpus,” HomeBank, 2016, https://homebank.talkbank.org/access/Password/Lyon.html

    2016

  26. [27]

    VanDam Cougar HomeBank Corpus,

    M. VanDam, “VanDam Cougar HomeBank Corpus,” Home- Bank, 2018, available at: https://homebank.talkbank.org/access/ Password/Cougar.html

    2018

  27. [28]

    Bergelson Seedlings HomeBank Corpus,

    E. Bergelson, “Bergelson Seedlings HomeBank Corpus,” Home- Bank, 2017, available at: https://homebank.talkbank.org/access/ Password/Bergelson.html

    2017

  28. [29]

    Long-form recordings from children in rossel island

    A. Cristia and M. Casillas, “Long-form recordings from children in rossel island.” 2020, unpublished raw data

    2020

  29. [30]

    The language 0-5 project,

    C. F. Rowland, S. Durrant, M. Peter, A. Bidgood, J. Pine, and L. S. Jago, “The language 0-5 project,” 2025. [Online]. Available: osf.io/kau5f

    2025

  30. [31]

    San Joaquin Valley HomeBank Corpus,

    A. S. Warlaumont, G. M. Pretzer, S. Mendoza, S. Schneider, J. Mutrie, L. Lopez, E. A. Walle, and C. T. Kello, “San Joaquin Valley HomeBank Corpus,” HomeBank, 2024, formerly the War- laumont HomeBank Corpus. Available at: https://homebank. talkbank.org/access/Password/SanJoaquin.html

    2024

  31. [32]

    Acoustical cues and grammatical units in speech to two preverbal infants,

    M. Soderstrom, M. Blossom, R. Foygel, and J. L. Morgan, “Acoustical cues and grammatical units in speech to two preverbal infants,”Journal of Child Language, vol. 35, no. 4, p. 869–902, 2008

    2008

  32. [33]

    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. 29, no. 2, pp. 196–215, 2024

    2024

  33. [34]

    PhonSES: A pilot study to measure socioeconomic status association with infants’ word and sound processing,

    A. Cristia, “PhonSES: A pilot study to measure socioeconomic status association with infants’ word and sound processing,” GIN,

  34. [35]

    Available: https://gin.g-node.org/LAAC-LSCP/ phonSES-public

    [Online]. Available: https://gin.g-node.org/LAAC-LSCP/ phonSES-public

  35. [36]

    Two-year-old chil- dren’s production of multiword utterances: A usage-based anal- ysis,

    E. Lieven, D. Salomo, and M. Tomasello, “Two-year-old chil- dren’s production of multiword utterances: A usage-based anal- ysis,”Cognitive Linguistics, vol. 20, no. 3, pp. 481–507, 2009

    2009

  36. [37]

    V ocal input and output among infants in a multilingual context: Evidence from long-form recordings in vanuatu,

    A. Cristia, L. Gautheron, and H. Colleran, “V ocal input and output among infants in a multilingual context: Evidence from long-form recordings in vanuatu,”Developmental Science, vol. 26, no. 4, p. e13375, 2023

    2023

  37. [38]

    Casillas Home- Bank Corpus,

    M. Casillas, P. Brown, and S. C. Levinson, “Casillas Home- Bank Corpus,” HomeBank, 2017, available at: https://homebank. talkbank.org/access/Secure/Casillas.html

    2017

  38. [39]

    Winnipeg HomeBank Corpus,

    M. Soderstrom, “Winnipeg HomeBank Corpus,” HomeBank, 2016, https://homebank.talkbank.org/access/Password/Winnipeg. html

    2016

  39. [40]

    Early language ex- perience in a tseltal mayan village,

    M. Casillas, P. Brown, and S. C. Levinson, “Early language ex- perience in a tseltal mayan village,”Child Development, vol. 91, no. 5, pp. 1819–1835, 2020

    2020

  40. [41]

    Improving automatic speech recogni- tion performance for low-resource languages with self-supervised models,

    J. Zhao and W.-Q. Zhang, “Improving automatic speech recogni- tion performance for low-resource languages with self-supervised models,”IEEE Journal of Selected Topics in Signal Processing, vol. 16, no. 6, pp. 1227–1241, 2022

    2022

  41. [42]

    wav2vec 2.0: A framework for self-supervised learning of speech representa- tions,

    A. Baevski, Y . Zhou, A. Mohamed, and M. Auli, “wav2vec 2.0: A framework for self-supervised learning of speech representa- tions,” inAdvances in Neural Information Processing Systems, H. Larochelle, M. Ranzato, R. Hadsell, M. Balcan, and H. Lin, Eds., vol. 33. Curran Associates, Inc., 2020, pp. 12 449–12 460

    2020

  42. [43]

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

    W.-N. Hsu, B. Bolte, Y .-H. H. Tsai, K. Lakhotia, R. Salakhut- dinov, and A. Mohamed, “HuBERT: Self-supervised speech representation learning by masked prediction of hidden units,” IEEE/ACM Trans. Audio, Speech and Lang. Proc., vol. 29, p. 3451–3460, 2021

    2021

  43. [44]

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

    S. Chen, C. Wang, Z. Chen, Y . Wu, S. Liu, Z. Chen, J. Li, N. Kanda, T. Yoshioka, X. Xiao, J. Wu, L. Zhou, S. Ren, Y . Qian, Y . Qian, J. Wu, M. Zeng, X. Yu, and F. Wei, “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

  44. [45]

    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,” in 2015 International Conference on Acoustics, Speech and Signal Processing. IEEE, 2015, pp. 5206–5210

    2015

  45. [46]

    Unsupervised Cross-Lingual Representation Learning for Speech Recognition,

    A. Conneau, A. Baevski, R. Collobert, A. Mohamed, and M. Auli, “Unsupervised Cross-Lingual Representation Learning for Speech Recognition,” inInterspeech, 2021, pp. 2426–2430

    2021

  46. [47]

    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, pp. 1035–1039

    2023

  47. [48]

    Context-aware transformer trans- ducer for speech recognition,

    F.-J. Chang, J. Liu, M. Radfar, A. Mouchtaris, M. Omologo, A. Rastrow, and S. Kunzmann, “Context-aware transformer trans- ducer for speech recognition,” inAutomatic Speech Recognition and Understanding Workshop. IEEE, 2021, pp. 503–510

    2021

  48. [49]

    Dialoguernn: An attentive rnn for emotion de- tection in conversations,

    N. Majumder, S. Poria, D. Hazarika, R. Mihalcea, A. Gelbukh, and E. Cambria, “Dialoguernn: An attentive rnn for emotion de- tection in conversations,” inProceedings of the AAAI conference on artificial intelligence, vol. 33, no. 01, 2019, pp. 6818–6825

    2019

  49. [50]

    Improv- ing speaker diarization for naturalistic child-adult conversational interactions using contextual information,

    M. Kumar, S. H. Kim, C. Lord, and S. Narayanan, “Improv- ing speaker diarization for naturalistic child-adult conversational interactions using contextual information,”The Journal of the Acoustical Society of America, vol. 147, no. 2, pp. EL196–EL200, 2020

    2020

  50. [51]

    An Open-Source V oice Type Classifier for Child-Centered Day- long Recordings,

    M. Lavechin, R. Bousbib, H. Bredin, E. Dupoux, and A. Cristia, “An Open-Source V oice Type Classifier for Child-Centered Day- long Recordings,” inInterspeech, 2020, pp. 3072–3076

    2020