Pith. sign in

REVIEW 3 major objections 5 minor 53 references

CAtCh: Cognitive Assessment through Cookie Thief

T0 review · 3 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 Acoustic and prosody features outperform linguistic ones for predicting cognitive impairment from one-minute Cookie Thief recordings.

desk verdict A careful, honest benchmark that translates ADRD and sentiment-analysis speech methods to broader CI prediction; its acoustic-beats-linguistic finding is worth taking seriously but is partly confounded by ASR and task transfer, and the abstract oversells multimodal superiority. read the letter →

arxiv 2506.06603 v1 pith:C23ABAL6 submitted 2025-06-07 cs.LG cs.AIcs.SDeess.AS

Joseph T Colonel , Carolyn Hagler , Guiselle Wismer , Laura Curtis , Jacqueline Becker , Juan Wisnivesky , Alex Federman , Gaurav Pandey This is my paper
classification cs.LGcs.AIcs.SDeess.AS
keywords cognitiveimpairmentspeechbiomarkersmultimodalmachinelearningacousticfeaturesprosodyCookieThiefTestnaturallanguageprocessingscreening
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

The paper evaluates six open-source machine-learning methods, originally built for dementia detection or multimodal sentiment analysis, on the task of predicting broad cognitive impairment from one-minute Cookie Thief picture descriptions. Its central claim is that acoustic and prosodic features, especially interpretable affect and prosody measures, carry more predictive signal than word-level linguistic features, including those from transformer-based language models. Multimodal methods generally outperform unimodal ones, although the advantage is not always statistically significant and one method actually loses accuracy when language is added. If correct, audio-only screening for cognitive impairment in primary care becomes a more realistic goal, with the key signal residing in how people speak rather than what they say.

What carries the argument

The comparison rests on paired unimodal and multimodal classifiers and on ablations that remove one modality at a time. The central feature families are interpretable acoustic features, including the IS10 paralinguistic set designed for age, gender, and affect, and prosody features that capture rate, silence, and pitch variation, versus linguistic features that range from expert-defined parts-of-speech and syntactic complexity to learned transformer-based text representations. Each classifier is trained and tested over 100 stratified train-test splits with minority-class oversampling, and pairwise significance is assessed with a Friedman test and Nemenyi correction. The ablations isolate which modality actually drives performance, and they are what expose the acoustic-dominance result.

What would settle it

A replication on a larger cohort with clinically confirmed cognitive-impairment diagnoses, in which interpretable prosody and affect features no longer significantly outperform linguistic features, or in which the best method's AUC does not exceed chance, would refute the paper's central claims.

Watch

Extended reading notes

Core claim

On a cohort of 157 older adults, the authors find that the best methods for classifying cognitive impairment from one-minute Cookie Thief recordings reach a mean area under the ROC curve of about 0.672 for a multimodal fusion of speech representations and paralinguistic features. The strongest unimodal classifiers use interpretable acoustic feature sets that capture affect and prosody, such as the IS10 paralinguistic challenge features and prosody features measuring speaking rate, silence, and pitch variation. These acoustic classifiers significantly outperform both transformer-based linguistic representations and expert-defined lexical and syntactic features. The authors conclude that, for broader cognitive impairment, acoustic and prosody cues dominate linguistic content, a reversal of the typical finding in Alzheimer's-specific speech detection.

Load-bearing premise

The study labels participants as cognitively impaired when their age- and education-adjusted score on a standard cognitive screening test falls more than one standard deviation below the norm; if that threshold misclassifies people, every method comparison inherits the error.

Editorial extensions

If this is right

  • Speech-based cognitive-impairment screening from a one-minute picture-description task is feasible with open-source methods, but current best performance (mean AUC around 0.67) is not yet clinical-grade.
  • Future screening tools should weight acoustic and prosody features at least as heavily as language content, since the former carried more signal in this cohort.
  • Unimodal baselines should be reported alongside multimodal models, because adding language to one method actually degraded accuracy relative to its prosody-only version.
  • Fine-tuning large pretrained models on small clinical cohorts gives only weak and inconsistent gains, so simpler feature-based classifiers remain competitive.
  • Methods developed for Alzheimer's dementia detection do not automatically transfer to broader cognitive impairment, so CI-specific evaluation is necessary.

Reading between the lines

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

  • The acoustic advantage may reflect that broad cognitive impairment affects motor speech and prosody earlier or more consistently than lexical and syntactic content, which could explain why language features tuned to Alzheimer's fail to transfer; testing this would require correlating specific prosody measures with neuropsychological subdomains.
  • Because prosody and affect features are largely language-independent, an audio-only CI screener built on them may generalize across languages and dialects without retraining, a testable extension the paper leaves implicit.
  • The multimodal methods that operated at the sentence level were averaged to patient level; a patient-level pooling or attention layer could plausibly improve on the reported mean AUC, a modification the paper does not explore.
  • The study's reliance on a single screening-threshold label suggests that a re-analysis with clinically adjudicated diagnoses could change the acoustic-versus-linguistic ranking; this is the natural next validation step.
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 / 5 minor

Summary. The paper evaluates six open-source machine learning methods developed for Alzheimer's disease and related dementia (ADRD) detection and for multimodal sentiment analysis, applying them to the new task of predicting broader cognitive impairment (CI) from one-minute Cookie Thief Test (CTT) audio recordings. The study uses a cohort of 157 primary-care patients (28 CI, 129 healthy), with CI defined by an age- and education-adjusted MoCA z-score below -1.0. Each method is evaluated over 100 stratified train/test splits with AUC and Fmax(CI) as metrics, using Friedman tests with Nemenyi post-hoc correction. The paper reports that multimodal methods outperformed unimodal ones, that acoustic features outperform linguistic features, and that finetuning of large pre-trained models gives weak support for CI prediction. The code is made available.

Significance. If the findings hold, the paper provides a useful benchmark for automated CI screening from short spontaneous speech, a clinically relevant and relatively underexplored task compared to ADRD. The study is notable for its systematic evaluation of several open-source methods, the use of repeated resampling with standard error reporting, appropriate non-parametric multiple-comparison statistics, and a public code repository. However, the central claims about multimodal and acoustic superiority are partially contradicted by the paper's own results and are confounded by the provenance of linguistic features; therefore the contribution is an instructive evaluation whose headline conclusions require substantial qualification.

major comments (3)
  1. [Abstract and Section V (Results)] The claim that 'multimodal methods outperformed unimodal ones' is not supported by the reported results. Figure 5 shows that Farzana et al.'s unimodal acoustic classifier significantly outperformed its multimodal version in both AUC and Fmax(CI) (p<0.001), and Figure 4 shows no significant difference between MFN's multimodal and unimodal configurations (p>0.9). Additionally, the unimodal acoustic method of Chen et al. significantly outperformed Farzana et al.'s multimodal method in Fmax(CI) (p=0.013). The abstract and discussion should be revised to state that only some multimodal methods (MISA, Ying et al.) outperformed some unimodal baselines, not that multimodal methods generally outperformed unimodal ones.
  2. [Section VI.C and Section II] The acoustic-versus-linguistic comparison is confounded by feature provenance. All linguistic features are computed from Whisper ASR-generated transcripts, while acoustic features are computed from the original audio; if ASR errors are more frequent among CI participants due to disfluencies or atypical prosody, linguistic features will be systematically degraded. In addition, the BERT model in Ying et al. was fine-tuned on ADReSSo-2021 ADRD transcripts, and the hand-crafted NLP features in Heitz and Farzana were selected specifically for ADRD detection; neither was adapted to the CI task or this multilingual cohort. Section VI.C acknowledges this possibility only speculatively and never tests it, so the abstract's assertion that 'interpretable acoustic features ... significantly outperform BERT-based linguistic features and interpretable linguistic features' remains under-supported. The authors should either run a control analysis (e.g., using manual transcripts or a CI-tuned language model) or substantially soften this conclusion.
  3. [Section II] The CI ground-truth labels are defined solely as a MoCA z-score below -1.0, which is not a clinical diagnosis of cognitive impairment. With only 28 CI participants, label noise from this threshold could substantially affect all method comparisons, including the acoustic-versus-linguistic comparison. The authors should report a sensitivity analysis using more conservative or more stringent thresholds (e.g., z < -1.5 or z < -0.5) or otherwise discuss how plausible label misclassification would impact the observed rankings and significance tests.
minor comments (5)
  1. [Section VI.A] The corpus description states '29/150 CI/control' but Section II reports 28 CI and 129 healthy participants; this inconsistency should be corrected.
  2. [Section V] The sentence 'To guide our exploration of these results, we structureQ2 into the following two sub-questions' appears twice; remove the duplicate.
  3. [Figure 4 caption] There is a typo in the caption: 'statistically singificant' should be 'statistically significant'.
  4. [Abstract and Section IV] The metric 'Fmax (CI)' is not defined before its first use in the abstract; please define it in the introduction or in a footnote.
  5. [Section IV] The oversampling ratio and the validation split fraction (75/25) are fixed; reporting these values explicitly in the text would improve reproducibility, as they are free parameters that could affect the results.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: all models are trained and evaluated on disjoint held-out splits; comparisons are empirical benchmarks, not self-referential definitions.

full rationale

The paper is an evaluation study: it imports six open-source ADRD/MSA methods, trains them on 75/25 stratified splits of the CTT corpus, and reports held-out AUC and Fmax averaged over 100 splits. The central claim (acoustic features outperform linguistic features for CI) is an empirical comparison of pre-existing feature sets, not a quantity defined by the method that produced it. The CI label comes from MoCA z-scores, a clinical instrument independent of the speech features. Method-specific hyperparameters from prior ADRD papers are reused, but this is a standard transfer-evaluation design and does not make the target result an input. The only self-citations are to the clinical study that supplied the corpus ([2]) and to a prior audio-processing paper by one author ([6]); neither supplies the fitted values or the conclusion. The discussion's caveat that English-specific NLP features may not transfer to a multilingual cohort is a validity concern, not circularity. No step reduces to its own input.

Assumptions & free parameters 3 free parameters · 6 assumptions · 0 invented entities

The central evaluation rests on the MoCA-based CI label, the assumption that Cookie Thief speech encodes CI, and several preprocessing choices (oversampling, segment length, validation split). None of these are fitted to the test data, but they are hand-chosen and could affect the reported ranking.

free parameters (3)
  • Oversampling ratio = 1:1 minority:majority
    Chosen by the authors to balance classes in training (Section IV); the central comparisons could shift with a different ratio.
  • Audio segment length and hop for acoustic DL models = 10 seconds with 2-second hop
    Carried over from Ying et al and Chen et al; not tuned for CI, and the acoustic advantage may depend on this chunking choice.
  • Validation split fraction = 0.25 of the training set
    Used for early stopping in MISA, Chen et al, and Ying et al (Section IV); another arbitrary split choice that can affect model selection.
assumptions (6)
  • domain assumption MoCA age- and education-adjusted z-score below -1.0 defines cognitive impairment.
    Section II uses this threshold to label 28 CI cases; if the MoCA screener is not a valid ground truth for CI in this primary-care cohort, all performance comparisons are affected.
  • domain assumption Cookie Thief Test audio and ASR transcripts contain sufficient signal for CI prediction.
    The entire evaluation presupposes that one-minute picture descriptions encode cognitive status; this is the premise of the study.
  • domain assumption Whisper ASR transcripts are accurate enough for linguistic feature extraction.
    Section II notes a manual review found no harmful hallucinations, but recognition errors could bias BERT and NLP features.
  • domain assumption Random oversampling of the minority CI class in the training split does not distort the comparison.
    Section IV oversamples to equal representation; test splits remain imbalanced, but the training distribution change could affect AUC and Fmax.
  • standard math Friedman test with Nemenyi post-hoc is valid for comparing 100 overlapping train/test splits.
    Section IV uses this standard procedure, but the splits are not independent, which may affect the reported p-values.
  • domain assumption The selected open-source repositories faithfully represent their original methods.
    Section III selects on code availability and manual inspection; implementation differences could change conclusions.

how reviews work

0 comments
Cite this review

Pith. "Pith review of CAtCh: Cognitive Assessment through Cookie Thief." pith.science (2026). https://pith.science/paper/C23ABAL6

@misc{pith2026250606603,
  author       = {Pith},
  title        = {Pith review of: CAtCh: Cognitive Assessment through Cookie Thief},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/C23ABAL6}},
  note         = {Machine review of arXiv:2506.06603}
}
read the original abstract

Several machine learning algorithms have been developed for the prediction of Alzheimer's disease and related dementia (ADRD) from spontaneous speech. However, none of these algorithms have been translated for the prediction of broader cognitive impairment (CI), which in some cases is a precursor and risk factor of ADRD. In this paper, we evaluated several speech-based open-source methods originally proposed for the prediction of ADRD, as well as methods from multimodal sentiment analysis for the task of predicting CI from patient audio recordings. Results demonstrated that multimodal methods outperformed unimodal ones for CI prediction, and that acoustics-based approaches performed better than linguistics-based ones. Specifically, interpretable acoustic features relating to affect and prosody were found to significantly outperform BERT-based linguistic features and interpretable linguistic features, respectively. All the code developed for this study is available at https://github.com/JTColonel/catch.

Figures

Figures reproduced from arXiv: 2506.06603 by the authors.

Figure 1
Figure 1. Demographics of the cohort used in this study. [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Google Scholar method search performed on August 13, 2024 for [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Mean AUC and Fmax (CI) per evaluated method over 100 train-test splits. Error bars show the standard error of the mean. MISA achieved the highest [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Mean AUC and Fmax (CI) for the MFN classifiers over 100 train-test splits. Error bars represent the standard error of the mean. The proposed [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Mean AUC and Fmax (CI) for Farzana et. al. classifiers over 100 train-test splits. Error bars represent the standard error of the mean. The acoustic [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: Mean AUC and Fmax(CI) for various Ying et. al. configurations over 100 train-test splits. Error bars represent the standard error of the mean. [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

53 extracted references · 41 canonical work pages

  1. [1]

    2022 alzheimer’s disease facts and figures,

    “2022 alzheimer’s disease facts and figures,”Alzheimer’s & Dementia, vol. 18, no. 4, pp. 700–789, 2022. [Online]. Available: https://alz-journals.onlinelibrary.wiley.com/doi/abs/10.1002/alz.12638

    work page doi:10.1002/alz.12638 2022

  2. [2]

    Rates of undiagnosed cognitive impairment and performance on the montreal cognitive assessment among older adults in primary care,

    A. D. Federman, J. H. Becker, M. R. Mindt, D. Cho, L. Curtis, and J. Wisnivesky, “Rates of undiagnosed cognitive impairment and performance on the montreal cognitive assessment among older adults in primary care,”Journal of general internal medicine, vol. 38, no. 11, pp. 2511–2518, 2023

    work page 2023

  3. [3]

    Brain health: the importance of recognizing cognitive impairment: an iagg consensus conference,

    J. E. Morley, J. C. Morris, M. Berg-Weger, S. Borson, B. D. Carpenter, N. Del Campo, B. Dubois, K. Fargo, L. J. Fitten, J. H. Flahertyet al., “Brain health: the importance of recognizing cognitive impairment: an iagg consensus conference,”Journal of the American Medical Directors Association, vol. 16, no. 9, pp. 731–739, 2015

    work page 2015

  4. [4]

    Cognitive screening for dementia in primary care: a systematic review,

    J. E. Yokomizo, S. S. Simon, and C. M. de Campos Bottino, “Cognitive screening for dementia in primary care: a systematic review,”Interna- tional psychogeriatrics, vol. 26, no. 11, pp. 1783–1804, 2014

    work page 2014

  5. [5]

    Underdiagnosis of dementia: an observational study of patterns in diagnosis and awareness in us older adults,

    H. Amjad, D. L. Roth, O. C. Sheehan, C. G. Lyketsos, J. L. Wolff, and Q. M. Samus, “Underdiagnosis of dementia: an observational study of patterns in diagnosis and awareness in us older adults,”Journal of general internal medicine, vol. 33, pp. 1131–1138, 2018

    work page 2018

  6. [6]

    Can you hear me now? clinical applications of audio recordings,

    A. Kumar, T. Jaquenoud, J. H. Becker, D. Cho, M. R. Mindt, A. Fed- erman, and G. Pandey, “Can you hear me now? clinical applications of audio recordings,”medRxiv, pp. 2022–02, 2022

    work page 2022

  7. [7]

    A systematic review and narrative analysis of digital speech biomarkers in motor neuron disease,

    M. Bowden, E. Beswick, J. Tam, D. Perry, A. Smith, J. Newton, S. Chandran, O. Watts, and S. Pal, “A systematic review and narrative analysis of digital speech biomarkers in motor neuron disease,”NPJ digital medicine, vol. 6, no. 1, p. 228, 2023

    work page 2023

  8. [8]

    Speech as a promising biosignal in precision psychiatry,

    M. Kappen, M.-A. Vanderhasselt, and G. M. Slavich, “Speech as a promising biosignal in precision psychiatry,”Neuroscience & Biobehav- ioral Reviews, vol. 148, p. 105121, 2023

    work page 2023

Show all 53 references
  1. [9]

    Artificial intelligence, speech, and language processing approaches to monitoring alzheimer’s disease: a systematic review,

    S. de la Fuente Garcia, C. W. Ritchie, and S. Luz, “Artificial intelligence, speech, and language processing approaches to monitoring alzheimer’s disease: a systematic review,”Journal of Alzheimer’s Disease, vol. 78, no. 4, pp. 1547–1574, 2020

    2020

  2. [10]

    Alexa, siri, cortana, and more: an introduction to voice assistants,

    M. B. Hoy, “Alexa, siri, cortana, and more: an introduction to voice assistants,”Medical reference services quarterly, vol. 37, no. 1, pp. 81– 88, 2018

    2018

  3. [11]

    Performance on the boston cookie theft picture description task in patients with early dementia of the alzheimer’s type: missing information,

    E. Giles, K. Patterson, and J. R. Hodges, “Performance on the boston cookie theft picture description task in patients with early dementia of the alzheimer’s type: missing information,”Aphasiology, vol. 10, no. 4, pp. 395–408, 1996

    1996

  4. [12]

    Detecting cognitive decline using speech only: The adresso challenge,

    S. Luz, F. Haider, S. de la Fuente, D. Fromm, and B. MacWhinney, “Detecting cognitive decline using speech only: The adresso challenge,” inINTERSPEECH 2021. ISCA, 2021

    2021

  5. [13]

    Multi- modal sentiment analysis: A systematic review of history, datasets, mul- timodal fusion methods, applications, challenges and future directions,

    A. Gandhi, K. Adhvaryu, S. Poria, E. Cambria, and A. Hussain, “Multi- modal sentiment analysis: A systematic review of history, datasets, mul- timodal fusion methods, applications, challenges and future directions,” Information Fusion, vol. 91, pp. 424–444, 2023

    2023

  6. [14]

    The distress analysis interview corpus of human and computer interviews

    J. Gratch, R. Artstein, G. M. Lucas, G. Stratou, S. Scherer, A. Nazarian, R. Wood, J. Boberg, D. DeVault, S. Marsellaet al., “The distress analysis interview corpus of human and computer interviews.” inLREC. Reykjavik, 2014, pp. 3123–3128

    2014

  7. [15]

    Automated assessment of psychiatric disorders using speech: A systematic review,

    D. M. Low, K. H. Bentley, and S. S. Ghosh, “Automated assessment of psychiatric disorders using speech: A systematic review,”Laryngoscope investigative otolaryngology, vol. 5, no. 1, pp. 96–116, 2020

    2020

  8. [16]

    Avec 2019 workshop and challenge: state-of-mind, detecting depression with ai, and cross-cultural affect recognition,

    F. Ringeval, B. Schuller, M. Valstar, N. Cummins, R. Cowie, L. Tavabi, M. Schmitt, S. Alisamir, S. Amiriparian, E.-M. Messneret al., “Avec 2019 workshop and challenge: state-of-mind, detecting depression with ai, and cross-cultural affect recognition,” inProceedings of the 9th...

    2019

  9. [17]

    Multilingual speech and language analysis for the assessment of mild cognitive impairment: Outcomes from the taukadial challenge,

    P. A. Pérez-Toro, T. Arias-Vergara, P. Klumpp, T. Weise, M. Schuster, E. Noeth, J. R. Orozco-Arroyave, and A. Maier, “Multilingual speech and language analysis for the assessment of mild cognitive impairment: Outcomes from the taukadial challenge,”Proc. Interspeech 2024, pp. 9...

    2024

  10. [18]

    The montreal cognitive assessment, moca: a brief screening tool for mild cognitive impairment,

    Z. S. Nasreddine, N. A. Phillips, V . Bédirian, S. Charbonneau, V . White- head, I. Collin, J. L. Cummings, and H. Chertkow, “The montreal cognitive assessment, moca: a brief screening tool for mild cognitive impairment,”Journal of the American Geriatrics Society, vol. 53, no....

    2005

  11. [19]

    Normative data for the montreal cognitive assessment (moca) in a population-based sample,

    H. C. Rossetti, L. H. Lacritz, C. M. Cullum, and M. F. Weiner, “Normative data for the montreal cognitive assessment (moca) in a population-based sample,”Neurology, vol. 77, no. 13, pp. 1272–1275, 2011

    2011

  12. [20]

    Stealing cookies in the twenty- first century: Measures of spoken narrative in healthy versus speakers with aphasia,

    S. Berube, J. Nonnemacher, C. Demsky, S. Glenn, S. Saxena, A. Wright, D. C. Tippett, and A. E. Hillis, “Stealing cookies in the twenty- first century: Measures of spoken narrative in healthy versus speakers with aphasia,”American journal of speech-language pathology, vol. 28, ...

    2019

  13. [21]

    Robust speech recognition via large-scale weak super- vision,

    A. Radford, J. W. Kim, T. Xu, G. Brockman, C. McLeavey, and I. Sutskever, “Robust speech recognition via large-scale weak super- vision,” inInternational Conference on Machine Learning. PMLR, 2023, pp. 28 492–28 518

    2023

  14. [22]

    whisper-timestamped,

    J. Louradour, “whisper-timestamped,” https://github.com/linto-ai/ whisper-timestamped, 2023

    2023

  15. [23]

    Automated evaluation of standardized dementia screening tests,

    F. Braun, M. Foerstel, B. Oppermann, A. Erzigkeit, T. Hillemacher, H. Lehfeld, and K. Riedhammer, “Automated evaluation of standardized dementia screening tests,”INTERSPEECH 2022 INTERSPEECH, vol. 2022, pp. 2478–2482, 2022

    2022

  16. [24]

    Misa: Modality-invariant and-specific representations for multimodal sentiment analysis,

    D. Hazarika, R. Zimmermann, and S. Poria, “Misa: Modality-invariant and-specific representations for multimodal sentiment analysis,” inPro- ceedings of the 28th ACM international conference on multimedia, 2020, pp. 1122–1131

    2020

  17. [25]

    Memory fusion network for multi-view sequential learning,

    A. Zadeh, P. P. Liang, N. Mazumder, S. Poria, E. Cambria, and L.-P. Morency, “Memory fusion network for multi-view sequential learning,” inProceedings of the AAAI conference on artificial intelligence, vol. 32, no. 1, 2018

    2018

  18. [26]

    Careless whisper: Speech-to-text hallucination harms,

    A. Koenecke, A. S. G. Choi, K. X. Mei, H. Schellmann, and M. Sloane, “Careless whisper: Speech-to-text hallucination harms,” inThe 2024 ACM Conference on Fairness, Accountability, and Transparency, 2024, pp. 1672–1681

    2024

  19. [27]

    The influence of automatic speech recognition on linguistic features and automatic alzheimer’s disease detection from spontaneous speech,

    J. Heitz, G. Schneider, and N. Langer, “The influence of automatic speech recognition on linguistic features and automatic alzheimer’s disease detection from spontaneous speech,” inProceedings of the 2024 Joint International Conference on Computational Linguistics, Language Re...

    2024

  20. [28]

    Speech and language processing: An introduction to natural language processing, computational linguistics, and speech recognition

    D. Jurafsky and J. H. Martin, “Speech and language processing: An introduction to natural language processing, computational linguistics, and speech recognition.”

  21. [29]

    Exploring multi- task learning and data augmentation in dementia detection with self- supervised pretrained models,

    M. Chen, C. Miao, J. Ma, S. Wang, and J. Xiao, “Exploring multi- task learning and data augmentation in dementia detection with self- supervised pretrained models,” inProc. INTERSPEECH, vol. 2023, 2023, pp. 5037–5041

    2023

  22. [30]

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

    W.-N. Hsu, B. Bolte, Y .-H. H. Tsai, K. Lakhotia, R. Salakhutdinov, and A. Mohamed, “Hubert: Self-supervised speech representation learning by masked prediction of hidden units,”IEEE/ACM transactions on audio, speech, and language processing, vol. 29, pp. 3451–3460, 2021

    2021

  23. [31]

    Sep- 28k: A dataset for stuttering event detection from podcasts with people who stutter,

    C. Lea, V . Mitra, A. Joshi, S. Kajarekar, and J. P. Bigham, “Sep- 28k: A dataset for stuttering event detection from podcasts with people who stutter,” inICASSP 2021-2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2021, pp. 6798–6802

    2021

  24. [32]

    Librispeech: an asr corpus based on public domain audio books,

    V . Panayotov, G. Chen, D. Povey, and S. Khudanpur, “Librispeech: an asr corpus based on public domain audio books,” in2015 IEEE international conference on acoustics, speech and signal processing (ICASSP). IEEE, 2015, pp. 5206–5210

    2015

  25. [33]

    Multimodal fusion for alzheimer’s disease recognition,

    Y . Ying, T. Yang, and H. Zhou, “Multimodal fusion for alzheimer’s disease recognition,”Applied Intelligence, vol. 53, no. 12, pp. 16 029– 16 040, 2023

    2023

  26. [34]

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

    A. Baevski, Y . Zhou, A. Mohamed, and M. Auli, “wav2vec 2.0: A framework for self-supervised learning of speech representations,” Advances in neural information processing systems, vol. 33, pp. 12 449– 12 460, 2020

    2020

  27. [35]

    Bert: Pre-training of deep bidirectional transformers for language understanding,

    J. Devlin, M.-W. Chang, K. Lee, and K. Toutanova, “Bert: Pre-training of deep bidirectional transformers for language understanding,”arXiv preprint arXiv:1810.04805, 2018

    2018 arXiv

  28. [36]

    The interspeech 2010 paralinguistic challenge,

    B. Schuller, S. Steidl, A. Batliner, F. Burkhardt, L. Devillers, C. Müller, and S. Narayanan, “The interspeech 2010 paralinguistic challenge,” in Proc. INTERSPEECH 2010, Makuhari, Japan, 2010, pp. 2794–2797

    2010

  29. [37]

    Aligning books and movies: Towards story-like visual ex- planations by watching movies and reading books,

    Y . Zhu, “Aligning books and movies: Towards story-like visual ex- planations by watching movies and reading books,”arXiv preprint arXiv:1506.06724, 2015

    2015 arXiv

  30. [38]

    Scikit-learn: Machine learning in python,

    F. Pedregosa, G. Varoquaux, A. Gramfort, V . Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V . Dubourget al., “Scikit-learn: Machine learning in python,”the Journal of machine Learning research, vol. 12, pp. 2825–2830, 2011

    2011

  31. [39]

    Towards domain-agnostic and domain- adaptive dementia detection from spoken language,

    S. Farzana and N. Parde, “Towards domain-agnostic and domain- adaptive dementia detection from spoken language,” inProceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), 2023, pp. 11 965–11 978

    2023

  32. [40]

    Moving beyond ku ˇcera and francis: A critical evaluation of current word frequency norms and the introduction of a new and improved word frequency measure for american english,

    M. Brysbaert and B. New, “Moving beyond ku ˇcera and francis: A critical evaluation of current word frequency norms and the introduction of a new and improved word frequency measure for american english,” Behavior research methods, vol. 41, no. 4, pp. 977–990, 2009

    2009

  33. [41]

    Towards an automatic evaluation of the dysarthria level of patients with parkinson’s disease,

    J. C. Vásquez-Correa, J. Orozco-Arroyave, T. Bocklet, and E. Nöth, “Towards an automatic evaluation of the dysarthria level of patients with parkinson’s disease,”Journal of communication disorders, vol. 76, pp. 21–36, 2018

    2018

  34. [42]

    Co- varep—a collaborative voice analysis repository for speech technolo- gies,

    G. Degottex, J. Kane, T. Drugman, T. Raitio, and S. Scherer, “Co- varep—a collaborative voice analysis repository for speech technolo- gies,” in2014 ieee international conference on acoustics, speech and signal processing (icassp). IEEE, 2014, pp. 960–964

    2014

  35. [43]

    Glove: Global vectors for word representation,

    J. Pennington, R. Socher, and C. D. Manning, “Glove: Global vectors for word representation,” inProceedings of the 2014 conference on empirical methods in natural language processing (EMNLP), 2014, pp. 1532–1543

    2014

  36. [44]

    A large- scale evaluation of computational protein function prediction,

    P. Radivojac, W. T. Clark, T. R. Oron, A. M. Schnoes, T. Wittkop, A. Sokolov, K. Graim, C. Funk, K. Verspoor, A. Ben-Huret al., “A large- scale evaluation of computational protein function prediction,”Nature methods, vol. 10, no. 3, pp. 221–227, 2013

    2013

  37. [45]

    Statistical comparisons of classifiers over multiple data sets,

    J. Demšar, “Statistical comparisons of classifiers over multiple data sets,” The Journal of Machine learning research, vol. 7, pp. 1–30, 2006

    2006

  38. [46]

    Hique: Hierarchical question embedding network for multimodal depression detection,

    J. Jung, C. Kang, J. Yoon, S. Kim, and J. Han, “Hique: Hierarchical question embedding network for multimodal depression detection,” arXiv preprint arXiv:2408.03648, 2024

    2024 arXiv

  39. [47]

    Multimodal machine learning: A survey and taxonomy,

    T. Baltrušaitis, C. Ahuja, and L.-P. Morency, “Multimodal machine learning: A survey and taxonomy,”IEEE transactions on pattern anal- ysis and machine intelligence, vol. 41, no. 2, pp. 423–443, 2018

    2018

  40. [48]

    Fine-tuning can distort pretrained features and underperform out-of-distribution,

    A. Kumar, A. Raghunathan, R. Jones, T. Ma, and P. Liang, “Fine-tuning can distort pretrained features and underperform out-of-distribution,” in International Conference on Learning Representations, 2022

    2022

  41. [49]

    Communicating emotion: The role of prosodic features

    R. W. Frick, “Communicating emotion: The role of prosodic features.” Psychological bulletin, vol. 97, no. 3, p. 412, 1985

    1985

  42. [50]

    Alzheimer’s dementia recognition through spontaneous speech,

    S. Luz, F. Haider, S. de la Fuente Garcia, D. Fromm, and B. MacWhin- ney, “Alzheimer’s dementia recognition through spontaneous speech,” p. 780169, 2021

    2021

  43. [51]

    Con- nected speech features from picture description in alzheimer’s disease: A systematic review,

    A. Slegers, R.-P. Filiou, M. Montembeault, and S. M. Brambati, “Con- nected speech features from picture description in alzheimer’s disease: A systematic review,”Journal of Alzheimer’s disease, vol. 65, no. 2, pp. 519–542, 2018

    2018

  44. [52]

    Who has undiagnosed dementia? a cross-sectional analysis of participants of the aging, demographics and memory study,

    G. M. Savva and A. Arthur, “Who has undiagnosed dementia? a cross-sectional analysis of participants of the aging, demographics and memory study,”Age and ageing, vol. 44, no. 4, pp. 642–647, 2015

    2015

  45. [53]

    Multilingual prediction of cognitive impair- ment with large language models and speech analysis,

    F. Agbavor and H. Liang, “Multilingual prediction of cognitive impair- ment with large language models and speech analysis,”Brain sciences, vol. 14, no. 12, p. 1292, 2024

    2024

Pith tools

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