Pith. sign in

REVIEW 4 major objections 6 minor 50 references

Comparative Evaluation of Acoustic Feature Extraction Tools for Clinical Speech Analysis

T0 review · 4 major / 6 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 features extracted from the same clinical speech recordings differ substantially across OpenSMILE, Praat, and Librosa, even when extraction parameters are standardized.

desk verdict A useful clinical-toolkit comparison whose central agreement claims are not supported because it uses Pearson r as an agreement metric, plus an unresolved sample-size discrepancy that must be fixed. read the letter →

arxiv 2506.01129 v2 pith:AMLCJVL5 submitted 2025-06-01 cs.SD eess.AS

Anna Seo Gyeong Choi , Alexander Richardson , Ryan Partlan , Sunny Tang , Sunghye Cho This is my paper · ORCID
classification cs.SDeess.AS
keywords acousticfeatureextractionclinicalspeechanalysisschizophreniaspectrumdisordersOpenSMILEPraatLibrosacross-toolkitreproducibilitybiomarkers
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper asks whether three widely used acoustic feature extraction toolkits—OpenSMILE, Praat, and Librosa—produce consistent measurements when applied to the same clinical speech recordings. The authors standardized frame size, hop size, window function, sampling rate, and F0 search range across all tools, then compared feature values on speech from people with schizophrenia spectrum disorders and healthy controls. They found that some features, especially F0 percentiles, correlate strongly across toolkits, while others, like F0 standard deviation and formant values, can be poorly correlated or even negatively correlated. The paper argues that this toolkit-dependent variation undermines reproducibility in clinical speech analysis and that researchers should standardize protocols, cross-validate with multiple toolkits, and report extraction details transparently. If this is right, many published clinical speech findings may be sensitive to an unexamined software choice.

What carries the argument

The load-bearing machinery is the standardized extraction pipeline and the pairwise Pearson correlation between toolkit outputs. The authors align sampling rate (16 kHz), frame size (60 ms), hop size (10 ms), Hamming window, disabled pre-emphasis, and F0 search range (55–1000 Hz), then extract eGeMAPS-style statistics from OpenSMILE, Praat (via Parselmouth), and Librosa. Agreement for each acoustic feature is defined by Pearson r between every toolkit pair, and Fisher's r-to-z tests whether these correlations differ between groups; AUC from ROC curves is used to gauge each feature's diagnostic separation. The correlation matrix is the argument's backbone: it is what converts 'the tools computed different numbers' into 'the tools cannot be used interchangeably for certain features.'

What would settle it

Run the same standardized pipeline on a corpus with hand-verified F0 and formant values and compute Bland-Altman limits of agreement between each toolkit pair; large absolute discrepancies on features that show high Pearson correlation would show that correlation alone does not establish cross-toolkit reliability.

Watch

Extended reading notes

Core claim

The central discovery is that cross-toolkit agreement is feature-specific and cannot be assumed from parameter alignment alone. With identical configuration, OpenSMILE and Librosa show very high correlation for F0 percentiles (r=0.993–0.999 for SSD, r=0.962–0.993 for HC), and OpenSMILE–Praat agrees almost as well for these percentiles. Yet F0 standard deviation has a negative correlation between OpenSMILE and Librosa (r=-0.536 for SSD), and formants F1–F3 are inconsistent across tools. The paper also reports that correlation patterns differ between diagnostic groups, with statistically significant differences for F0 mean, HNR mean, and amplitude percentiles, and that in single-tool classification OpenSMILE-extracted F0 mean, HNR, and MFCC1 reach AUC above 0.75 for separating SSD from HC. These results are presented as evidence that toolkit choice is a source of uncontrolled variance in clinical speech analysis.

Load-bearing premise

The paper equates high Pearson correlation between two toolkits' outputs with cross-toolkit consistency, but Pearson r does not measure absolute agreement: one toolkit could be a scaled and offset version of another and still register near-perfect correlation.

Editorial extensions

If this is right

  • Studies that rely on a single toolkit for features like F0 standard deviation or formants may reach conclusions that another toolkit would not reproduce.
  • F0 percentiles are the most portable feature type across toolkits, making them a safer default for multi-site or longitudinal clinical studies.
  • Negative cross-toolkit correlations imply that combining features from different toolkits into one model could inject inverted signals, not merely additive noise.
  • Transparent reporting of toolkit names, versions, and parameter values becomes a necessary condition for reproducibility in clinical speech research.
  • Multi-toolkit cross-validation should be considered a standard step before any acoustic feature is used as a clinical marker.

Reading between the lines

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

  • The paper's consistency criterion is Pearson correlation, which tracks linear association rather than absolute agreement; a toolkit that returns a scaled and offset version of another tool's values would still show r near 1. Re-analyzing the same data with Bland-Altman limits of agreement or concordance correlation coefficients would test whether 'high correlation' actually means 'clinically inter
  • Because the paper computes correlations within each diagnostic group and finds group differences, an unstated implication is that toolkit agreement itself may be population-dependent; a feature validated on healthy controls might not transfer to a clinical group, and vice versa.
  • The paper's warning about deep embeddings masking extraction inconsistencies suggests a testable extension: feed the same audio to an embedding model and check whether downstream predictions vary with the front-end preprocessing toolkit, not just with the embedding weights.
  • A practical protocol implied by the findings is to report a per-feature 'toolkit consistency profile' before publishing clinical results, so that features with low cross-toolkit agreement are flagged as unreliable.
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 / 6 minor

Summary. The paper compares three acoustic feature extraction toolkits—OpenSMILE, Praat (via Parselmouth), and Librosa—applied to clinical speech recordings from individuals with schizophrenia spectrum disorders and healthy controls. After aligning extraction parameters across toolkits, the authors compute Pearson correlations between toolkit pairs for each feature, report group-wise correlation patterns, and evaluate the discriminative potential of selected features with ROC AUC. The central empirical claim is that cross-toolkit agreement is feature-dependent: F0 percentiles correlate highly across toolkits, while F0 standard deviation, formants, and several other measures correlate poorly or even negatively, leading to recommendations for standardized protocols, multi-toolkit validation, and transparent reporting in clinical speech analysis.

Significance. If the central claim is established, the paper addresses a genuine and practically important reproducibility problem: clinical speech researchers routinely use different feature extraction toolkits, and uncontrolled toolkit differences could explain contradictory findings across studies. The manuscript has clear strengths: it uses a real clinical dataset, attempts to standardize parameters across three widely used toolkits, and reports correlation results for a broad set of clinically relevant features. The recommendation to report tool versions, parameters, and multi-toolkit validation is sensible and actionable. However, the quantitative support for the central claim is weakened by an unresolved sample-size inconsistency between the abstract and the full text, the use of Pearson correlation as a proxy for agreement, post-hoc outlier trimming with an unclear rule, and in-sample AUC labeling without multiple-comparison control. These issues are fixable, but they currently leave the main feature-reliability dichotomy less well established than the text suggests.

major comments (4)
  1. [Abstract and §3.1] The abstract reports N=77 SSD and N=87 HC participants, whereas §3.1 and the rest of the full text report N=33 SSD and N=38 HC. Every correlation coefficient, Fisher z comparison, and AUC value in §4 depends on this N, so the inconsistency affects all quantitative results. Please correct the abstract and re-verify every statistic; if the larger numbers come from a different analysis or a different data release, state that explicitly.
  2. [§3.5 and §4] The paper uses Pearson correlation to 'assess agreement' and then partitions features into reliable (F0 percentiles) and unreliable (F0 SD, formants). Pearson r measures linear association, not absolute agreement: one toolkit can be an affine or nonlinear transform of another while still yielding r near 1. Conversely, a narrow participant-level range can keep r low even when absolute differences are small and clinically irrelevant. To support the reliable/unreliable dichotomy, report Bland-Altman limits of agreement, intraclass correlation, or mean absolute error, together with per-toolkit means and standard deviations. Without such metrics, the conclusion that F0 percentiles are interchangeable across toolkits is not established.
  3. [§4, Figure 1] The results are described as 'after removing outliers at the 25th and 75th percentiles.' This is not a standard outlier rule and is ambiguous: it could mean deleting values below the 25th and above the 75th percentile, using a 1.5×IQR fence, or winsorizing. Post-hoc trimming can substantially change Pearson correlations, especially in small samples. Please specify the exact rule, report how many observations were removed per feature and toolkit, and show whether the main conclusions hold without trimming.
  4. [§3.5 and §4] Many pairwise correlations are tested across features, toolkit pairs, and diagnostic groups, yet all p-values are reported against uncorrected thresholds. The Fisher's r-to-z comparisons mentioned in §3.5 are not reported in detail; §4 only states that differences were 'statistically significant (p<0.05) in features such as F0 mean, HNR mean, and amplitude percentiles.' Additionally, the AUC values in §4 are computed without any description of train/validation splitting or cross-validation, so they are in-sample and should be labeled as descriptive discrimination indices rather than predictive classification performance. Please add multiple-comparison control or clearly label the analysis as exploratory, and report the Fisher z results or remove that claim.
minor comments (6)
  1. [§4] The text says 'Amplitude mean shows strong correlation between OpenSMILE and Librosa ... but poorer agreement with Praat (r=-0.052 for SSD vs Librosa, p>0.05; r=0.425 for HC vs Librosa, p<0.001).' The phrase 'vs Librosa' is confusing; it appears to report the correlation between Praat and Librosa. Please clarify the notation consistently throughout.
  2. [Abstract and §4] The abstract says 'AUC greater than 0.70' for F0 mean, HNR, and MFCC1, while §4 says 'AUC values above 0.75, particularly when extracted using OpenSMILE.' Please reconcile these thresholds and report the actual AUC values with confidence intervals.
  3. [§3.3] The manuscript does not state the toolkit versions (e.g., OpenSMILE version, Praat/Parselmouth version, Librosa version) or the exact eGeMAPS version used. Since tool-version differences are a known source of variation, please report them for reproducibility.
  4. [Figure 1] The figure caption says empty cells indicate toolkit pairs not available for specific features, but the main text does not explain why some feature/toolkit pairs are unavailable. Please state the reason (e.g., a toolkit does not implement a given feature) in the caption or in §3.3.
  5. [§3.4] The sentence 'Amplitude captures the perceived intensity of speech' conflates an acoustic measure with a perceptual one. Consider saying 'intensity' or 'RMS energy' and noting the distinction from perceived loudness.
  6. [General] The manuscript does not include a data or code availability statement. For a reproducibility-focused comparison, providing the extraction pipeline and the extracted feature values would substantially strengthen the contribution.

Circularity Check

0 steps flagged · score 1.0 of 10

No circular derivation chain: the cross-toolkit comparison is empirical and self-contained, with only a minor non-load-bearing self-citation.

full rationale

This paper makes no derivational claim that could collapse into its own input. The central result is an empirical comparison: the same clinical recordings are fed to three toolkits under aligned parameters, and the reported Pearson correlations and AUC values are summary statistics of the observed outputs. No parameter is fitted to a subset and then renamed a prediction, and no equation defines a toolkit's feature in terms of the target result. The only self-citation found is reference [35] (Shellikeri et al., with co-author S. Cho), used in Section 3.4 solely to motivate the clinical relevance of formant ratios; it is not load-bearing for the comparison. The notable weaknesses are statistical rather than circular: Pearson r measures linear association rather than interchangeability (Sections 3.5 and 4), and the abstract's 77 SSD/87 HC contradicts Section 3.1's 33 SSD/38 HC, which is an internal-consistency/correctness issue. Neither weakness constitutes circularity. Given the incidental self-citation and the otherwise self-contained empirical analysis, a circularity score of 1 is appropriate.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The paper introduces no new theoretical entities. It relies on hand-chosen analysis thresholds and on standard but unexamined statistical assumptions about correlation-based agreement.

free parameters (2)
  • Outlier trimming thresholds = 25th and 75th percentiles
    Chosen by hand and applied before correlation analysis (Section 4, Figure 1). This removal alters correlation estimates and is not justified by a pre-registered criterion.
  • AUC threshold for 'good discrimination' = 0.70 (later 0.75)
    Used to label features as promising discriminators (Sections 3.5 and 4). Arbitrary and not corrected for multiple comparisons.
assumptions (4)
  • domain assumption Standardized extraction parameters make the three toolkits' outputs comparable, so residual differences reflect implementation differences rather than configuration differences.
    Section 3.3 lists common parameters but acknowledges 'certain toolkit-specific differences remained unavoidable.' The comparison still assumes the alignment is meaningful.
  • domain assumption Pearson correlation is an appropriate measure of agreement or reliability.
    Section 3.5 defines agreement as Pearson correlation; high r is interpreted as consistency, though r ignores systematic bias.
  • ad hoc to paper The outlier removal procedure does not materially bias the reported correlations.
    Section 4: 'after removing outliers at the 25th and 75th percentiles.' The exact procedure is not specified, and removing a large fraction of data can change correlation signs.
  • domain assumption Statistical significance at p<0.05 is reliable without correction for the large number of correlations tested.
    Section 3.5 reports p thresholds; dozens of correlations are tested per group, and no multiple-comparison correction is mentioned.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Comparative Evaluation of Acoustic Feature Extraction Tools for Clinical Speech Analysis." pith.science (2026). https://pith.science/paper/AMLCJVL5

@misc{pith2026250601129,
  author       = {Pith},
  title        = {Pith review of: Comparative Evaluation of Acoustic Feature Extraction Tools for Clinical Speech Analysis},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/AMLCJVL5}},
  note         = {Machine review of arXiv:2506.01129}
}
read the original abstract

This study compares three acoustic feature extraction toolkits (OpenSMILE, Praat, and Librosa) applied to clinical speech data from individuals with schizophrenia spectrum disorders (SSD) and healthy controls (HC). By standardizing extraction parameters across the toolkits, we analyzed speech samples from 77 SSD and 87 HC participants and found significant toolkit-dependent variations. While F0 percentiles showed high cross-toolkit correlation (r=0.962 to 0.999), measures like F0 standard deviation and formant values often had poor, even negative, agreement. Additionally, correlation patterns differed between SSD and HC groups. Classification analysis identified F0 mean, HNR, and MFCC1 (AUC greater than 0.70) as promising discriminators. These findings underscore reproducibility concerns and advocate for standardized protocols, multi-toolkit cross-validation, and transparent reporting.

Figures

Figures reproduced from arXiv: 2506.01129 by the authors.

Figure 1
Figure 1. Correlation heatmap comparing acoustic feature extraction across three tools (OpenSMILE [OS], Praat [PR], and Librosa [LR]) for SSD and HC groups. Color intensity indicates correlation strength from -1 (dark blue) to 1 (dark red). Statistical significance of correlations is marked: * (p<0.05), ** (p<0.01), and *** (p<0.001). Empty cells indicate toolkit pairs not available in specific features. extraction itself lac… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

50 extracted references · 48 canonical work pages

  1. [1]

    Although end-to-end neural architectures have been in- creasingly adopted, classical feature-based approaches are still widely used

    Introduction Acoustic features play a central role in speech processing pipelines, underpinning tasks such as speaker recognition [1, 2], emotion classification [3, 4], and clinical speech assessment [5, 6]. Although end-to-end neural architectures have been in- creasingly adopted, classical feature-based approaches are still widely used. In clinical cont...

    work page

  2. [2]

    Previous Studies The comparison and validation of acoustic feature extraction tools have been ongoing concerns in speech processing research [13, 14], yet systematic evaluations specifically targeting clin- ical applications remain limited. Early comparative studies across different toolkits focused primarily on general speech analysis [15, 16], leaving a...

    work page

  3. [3]

    Comparative Evaluation of Acoustic Feature Extraction Tools for Clinical Speech Analysis

    Methods 3.1. Data collection Participants diagnosed with schizophrenia spectrum disorders (SSD; N = 33, females = 24.7%, mean age = 35.92) were en- rolled from both inpatient and outpatient departments at a hos- pital in the US. Participants underwent screening using the psy- chosis and mood modules of the Structured Clinical Interview arXiv:2506.01129v2 ...

    work page Pith review arXiv 2025

  4. [4]

    Figure 1 presents the correlation matrix for both the SSD and HC groups after removing outliers at the 25th and 75th percentiles

    Results The correlation analysis reveals distinct patterns of agreement between the three feature extraction tools. Figure 1 presents the correlation matrix for both the SSD and HC groups after removing outliers at the 25th and 75th percentiles. For F0 percentile measurements, we observe remarkably high correlations, particularly between OpenSMILE and Li-...

    work page

  5. [5]

    These results have important implications for clinical speech analysis and highlight several critical considera- tions for future research

    Discussion & Conclusion Our findings reveal significant inconsistencies in feature extrac- tion results, despite careful parameter alignment and standard- ized processing. These results have important implications for clinical speech analysis and highlight several critical considera- tions for future research. The observed discrepancies in feature values ...

    work page

  6. [6]

    Development of standardized extraction protocols specifi- cally designed for clinical speech analysis, with validated pa- rameters across different toolkits

    work page

  7. [7]

    Increased transparency in research publications about extrac- tion methods, tool versions, and parameter configurations to improve reproducibility

    work page

  8. [8]

    Cross-validation of acoustic features using multiple extrac- tion tools and datasets before drawing clinical conclusions

    work page

Show all 50 references
  1. [9]

    Greater collaboration between speech technology experts, linguists, and clinical practitioners to ensure feature interpre- tation is grounded in both technical accuracy and clinical rel- evance

  2. [10]

    As automated speech anal- ysis continues to gain traction in healthcare applications, en- suring reliable, interpretable, and consistent measurements be- comes imperative

    Critical evaluation of newer embedding approaches in clin- ical contexts, with careful consideration of the trade-off be- tween performance and interpretability The field stands at a critical juncture where computational advances must be balanced against clinical needs for tra...

  3. [11]

    Selection of acoustic features for speaker identifica- tion,

    M. Sambur, “Selection of acoustic features for speaker identifica- tion,” IEEE Transactions on Acoustics, Speech, and Signal Pro- cessing, vol. 23, no. 2, pp. 176–182, 1975

    1975

  4. [12]

    Acoustic feature shuffling network for text-independent speaker verifica- tion

    J. Li, X. Fang, F. Chu, T. Gao, Y . Song, and R. L. Dai, “Acoustic feature shuffling network for text-independent speaker verifica- tion.” in Interspeech, 2022, pp. 4790–4794

    2022

  5. [13]

    Acoustic feature selection for automatic emotion recognition from speech,

    J. Rong, G. Li, and Y .-P. P. Chen, “Acoustic feature selection for automatic emotion recognition from speech,”Information pro- cessing & management, vol. 45, no. 3, pp. 315–328, 2009

    2009

  6. [14]

    Survey on bimodal speech emotion recognition from acoustic and linguistic information fu- sion,

    B. T. Atmaja, A. Sasou, and M. Akagi, “Survey on bimodal speech emotion recognition from acoustic and linguistic information fu- sion,” Speech Communication, vol. 140, pp. 11–28, 2022

    2022

  7. [15]

    Selection and combination of acoustic features for the description of pathologic voices,

    D. Michaelis, M. Fr ¨ohlich, and H. W. Strube, “Selection and combination of acoustic features for the description of pathologic voices,” The Journal of the Acoustical Society of America , vol. 103, no. 3, pp. 1628–1639, 1998

    1998

  8. [16]

    Machine learning approach for clas- sification of parkinson disease using acoustic features,

    V . Mittal and R. Sharma, “Machine learning approach for clas- sification of parkinson disease using acoustic features,” Journal of Reliable Intelligent Environments , vol. 7, no. 3, pp. 233–239, 2021

    2021

  9. [17]

    Multi- modal voice pathology detection architecture based on deep and handcrafted feature fusion,

    A. N. Omeroglu, H. M. Mohammed, and E. A. Oral, “Multi- modal voice pathology detection architecture based on deep and handcrafted feature fusion,”Engineering Science and Technology, an International Journal, vol. 36, p. 101148, 2022

    2022

  10. [18]

    Acoustic feature extraction with interpretable deep neural network for neu- rodegenerative related disorder classification,

    Y . Pan, B. Mirheidari, Z. Tu, R. O’Malley, T. Walker, A. Ven- neri, M. Reuber, D. Blackburn, and H. Christensen, “Acoustic feature extraction with interpretable deep neural network for neu- rodegenerative related disorder classification,” in Proceedings of Interspeech 2020. I...

    2020

  11. [19]

    Opensmile: the mu- nich versatile and fast open-source audio feature extractor,

    F. Eyben, M. W ¨ollmer, and B. Schuller, “Opensmile: the mu- nich versatile and fast open-source audio feature extractor,” in Proceedings of the 18th ACM international conference on Mul- timedia, 2010, pp. 1459–1462

    2010

  12. [20]

    Speak and unspeak with praat,

    P. Boersma and V . Van Heuven, “Speak and unspeak with praat,” Glot International, vol. 5, no. 9/10, pp. 341–347, 2001

    2001

  13. [21]

    Introducing parsel- mouth: A python interface to praat,

    Y . Jadoul, B. Thompson, and B. De Boer, “Introducing parsel- mouth: A python interface to praat,”Journal of Phonetics, vol. 71, pp. 1–15, 2018

    2018

  14. [22]

    librosa: Audio and music signal analysis in python

    B. McFee, C. Raffel, D. Liang, D. P. Ellis, M. McVicar, E. Bat- tenberg, and O. Nieto, “librosa: Audio and music signal analysis in python.” in SciPy, 2015, pp. 18–24

    2015

  15. [23]

    Techniques for feature extrac- tion in speech recognition system: A comparative study,

    U. Shrawankar and V . M. Thakare, “Techniques for feature extrac- tion in speech recognition system: A comparative study,” arXiv preprint arXiv:1305.1145, 2013

    2013 arXiv

  16. [24]

    A comparative study of feature extraction tech- niques for speech recognition system,

    P. K. Kurzekar, R. R. Deshmukh, V . B. Waghmare, and P. P. Shrishrimal, “A comparative study of feature extraction tech- niques for speech recognition system,” International Journal of Innovative Research in Science, Engineering and Technology , vol. 3, no. 12, pp. 18 006–18 016, 2014

    2014

  17. [25]

    Surfboard: Audio feature extraction for modern machine learning,

    R. Lenain, J. Weston, A. Shivkumar, and E. Fristed, “Surfboard: Audio feature extraction for modern machine learning,” arXiv preprint arXiv:2005.08848, 2020

    2005 arXiv

  18. [26]

    Speech acoustic (spac): A novel tool for speech feature extraction and classification,

    T. ¨Ozseven and M. D ¨u˘genci, “Speech acoustic (spac): A novel tool for speech feature extraction and classification,” Applied Acoustics, vol. 136, pp. 1–8, 2018

    2018

  19. [27]

    Repeatability of commonly used speech and lan- guage features for clinical applications,

    G. M. Stegmann, S. Hahn, J. Liss, J. Shefner, S. B. Rutkove, K. Kawabata, S. Bhandari, K. Shelton, C. J. Duncan, and V . Berisha, “Repeatability of commonly used speech and lan- guage features for clinical applications,” Digital biomarkers , vol. 4, no. 3, pp. 109–122, 2020

    2020

  20. [28]

    The reproducibility of bio- acoustic features is associated with sample duration, speech task, and gender,

    S. A. Almaghrabi, D. Thewlis, S. Thwaites, N. C. Rogasch, S. Lau, S. R. Clark, and M. Baumert, “The reproducibility of bio- acoustic features is associated with sample duration, speech task, and gender,” IEEE Transactions on Neural Systems and Rehabili- tation Engineering, vol...

    2022

  21. [29]

    Deep audio em- beddings for vocalisation clustering,

    P. Best, S. Paris, H. Glotin, and R. Marxer, “Deep audio em- beddings for vocalisation clustering,” Plos one, vol. 18, no. 7, p. e0283396, 2023

    2023

  22. [30]

    Com- bining deep embeddings of acoustic and articulatory features for speaker identification,

    Q.-B. Hong, C.-H. Wu, H.-M. Wang, and C.-L. Huang, “Com- bining deep embeddings of acoustic and articulatory features for speaker identification,” in ICASSP 2020-2020 IEEE Interna- tional Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2020, pp. 7589–7593

    2020

  23. [31]

    The structured clinical interview for DSM-IV axis I disorders (SCID-I) and the structured clinical in- terview for DSM-IV axis II disorders (SCID-II),

    M. B. First and M. Gibbon, “The structured clinical interview for DSM-IV axis I disorders (SCID-I) and the structured clinical in- terview for DSM-IV axis II disorders (SCID-II),” in Comprehen- sive Handbook of Psychological Assessment, Vol. 2. Personality Assessment, M. J. Hi...

    2004

  24. [32]

    Automatic speech emotion recognition: a systematic literature review,

    H. H. Mustafa, N. R. Darwish, and H. A. Hefny, “Automatic speech emotion recognition: a systematic literature review,” In- ternational Journal of Speech Technology, pp. 1–19, 2024

    2024

  25. [33]

    Identification of digital voice biomarkers for cognitive health,

    H. Lin, C. Karjadi, T. F. Ang, J. Prajakta, C. McManus, T. W. Alhanai, J. Glass, and R. Au, “Identification of digital voice biomarkers for cognitive health,” Exploration of medicine, vol. 1, p. 406, 2020

    2020

  26. [34]

    Paralin- guistic cues in speech to adapt robot behavior in human-robot in- teraction,

    A. Ashok, J. Pawlak, S. Paplu, Z. Zafar, and K. Berns, “Paralin- guistic cues in speech to adapt robot behavior in human-robot in- teraction,” in2022 9th IEEE RAS/EMBS International Conference for Biomedical Robotics and Biomechatronics (BioRob). IEEE, 2022, pp. 01–06

    2022

  27. [35]

    The geneva minimalistic acoustic parameter set (gemaps) for voice research and affective computing,

    F. Eyben, K. R. Scherer, B. W. Schuller, J. Sundberg, E. Andr ´e, C. Busso, L. Y . Devillers, J. Epps, P. Laukka, S. S. Narayanan et al., “The geneva minimalistic acoustic parameter set (gemaps) for voice research and affective computing,”IEEE transactions on affective computi...

    2015

  28. [36]

    Towards auto- matic assessment of voice disorders: A clinical approach

    P. Barche, K. Gurugubelli, and A. K. Vuppala, “Towards auto- matic assessment of voice disorders: A clinical approach.” in IN- TERSPEECH, 2020, pp. 2537–2541

    2020

  29. [37]

    Practical acoustics in clinical voice assessment: a praat primer,

    Y . Maryn, “Practical acoustics in clinical voice assessment: a praat primer,” Perspectives of the ASHA Special Interest Groups, vol. 2, no. 3, pp. 14–32, 2017

    2017

  30. [38]

    Using praat for linguistic research,

    W. Styler, “Using praat for linguistic research,” University of Col- orado at Boulder Phonetics Lab, 2013

    2013

  31. [39]

    Deep learning approach for sound sig- nal processing,

    M. Bhavya and M. Anala, “Deep learning approach for sound sig- nal processing,” in 2022 International Conference on Futuristic Technologies (INCOFT). IEEE, 2022, pp. 1–4

    2022

  32. [40]

    Evaluating the clinical utility of speech analysis and machine learning in schizophrenia: A pilot study,

    J. Huang, Y . Zhao, Z. Tian, W. Qu, X. Du, J. Zhang, Y . Tan, Z. Wang, and S. Tan, “Evaluating the clinical utility of speech analysis and machine learning in schizophrenia: A pilot study,” Computers in Biology and Medicine, vol. 164, p. 107359, 2023

    2023

  33. [41]

    Clinical classifica- tion of memory and cognitive impairment with multimodal digital biomarkers,

    R. Banks, C. Higgins, B. R. Greene, A. Jannati, J. Gomes-Osman, S. Tobyne, D. Bates, and A. Pascual-Leone, “Clinical classifica- tion of memory and cognitive impairment with multimodal digital biomarkers,” Alzheimer’s & Dementia: Diagnosis, Assessment & Disease Monitoring, vol...

    2024

  34. [42]

    V oice patterns as markers of schizophrenia: building a cumulative generalizable approach via a cross-linguistic and meta-analysis based investi- gation,

    A. Parola, A. Simonsen, J. M. Lin, Y . Zhou, H. Wang, S. Ubukata, K. Koelkebeck, V . Bliksted, and R. Fusaroli, “V oice patterns as markers of schizophrenia: building a cumulative generalizable approach via a cross-linguistic and meta-analysis based investi- gation,” Schizophr...

    2023

  35. [43]

    The aprosody of schizophrenia: Computationally derived acoustic phonetic underpinnings of monotone speech,

    M. T. Compton, A. Lunden, S. D. Cleary, L. Pauselli, Y . Alo- layan, B. Halpern, B. Broussard, A. Crisafio, L. Capulong, P. M. Balducci et al., “The aprosody of schizophrenia: Computationally derived acoustic phonetic underpinnings of monotone speech,” Schizophrenia research, ...

    2018

  36. [44]

    Initial evidence of vowel space reduction in a subset of individuals with schizophrenia,

    A. Hogoboom, M. Rouch, D. Lauerman, L. Pauselli, and M. T. Compton, “Initial evidence of vowel space reduction in a subset of individuals with schizophrenia,” Schizophrenia Research, vol. 255, pp. 158–164, 2023

    2023

  37. [45]

    Digital markers of motor speech impairments in spontaneous speech of patients with als-ftd spectrum disorders,

    S. Shellikeri, S. Cho, S. Ash, C. Gonzalez-Recober, C. T. McMil- lan, L. Elman, C. Quinn, D. A. Amado, M. Baer, D. J. Irwinet al., “Digital markers of motor speech impairments in spontaneous speech of patients with als-ftd spectrum disorders,” Amyotrophic Lateral Sclerosis and...

    2024

  38. [46]

    V ocal acoustic features may be objective biomarkers of negative symptoms in schizophrenia: A cross-sectional study,

    Q. Zhao, W.-Q. Wang, H.-Z. Fan, D. Li, Y .-J. Li, Y .-L. Zhao, Z.- X. Tian, Z.-R. Wang, Y .-L. Tan, and S.-P. Tan, “V ocal acoustic features may be objective biomarkers of negative symptoms in schizophrenia: A cross-sectional study,” Schizophrenia Research, vol. 250, pp. 180–185, 2022

    2022

  39. [47]

    Algorithm for jitter and shimmer measurement in pathologic voices,

    J. P. Teixeira and A. Gonc ¸alves, “Algorithm for jitter and shimmer measurement in pathologic voices,” Procedia Computer Science, vol. 100, pp. 271–279, 2016

    2016

  40. [48]

    V ocal jitter and shimmer in stuttering,

    P. W. Newman, R. W. Harris, and L. M. Hilton, “V ocal jitter and shimmer in stuttering,”Journal of fluency disorders, vol. 14, no. 2, pp. 87–95, 1989

    1989

  41. [49]

    Prediction of negative symptoms of schizophrenia from emotion related low-level speech signals,

    D. Chakraborty, Z. Yang, Y . Tahir, T. Maszczyk, J. Dauwels, N. Thalmann, J. Zheng et al., “Prediction of negative symptoms of schizophrenia from emotion related low-level speech signals,” in 2018 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP...

    2018

  42. [50]

    Clinical investigation of speech signal fea- tures among patients with schizophrenia,

    J. Zhang, P. A. N. Zhongde, G. U. I. Chao, Z. H. U. Jie, and C. U. I. Donghong, “Clinical investigation of speech signal fea- tures among patients with schizophrenia,” Shanghai Archives of Psychiatry, vol. 28, no. 2, p. 95, 2016

    2016

Pith tools

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