pith. sign in

arxiv: 2606.21411 · v1 · pith:GHIV37QAnew · submitted 2026-06-19 · 💻 cs.SD

CoughPhase-CLR: Designing an acoustics-informed foundation model for coughing sound classification

Marius Moldovan , Anton Batliner , Thomas M. Berghaus , Bj\"orn W. Schuller , Andreas Triantafyllopoulos This is my paper

Pith reviewed 2026-06-26 12:54 UTC · model grok-4.3

classification 💻 cs.SD
keywords contrastive learningself-supervised learningcough classificationrespiratory soundsCOVID-19 detectionCOPD classificationacoustic phasesfoundation model
0
0 comments X

The pith

CoughPhase-CLR pairs audio segments from the same cough phase for contrastive pre-training, outperforming random cropping on health classification tasks.

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

The paper introduces CoughPhase-CLR, a self-supervised framework that builds positive pairs from the physiological phases of cough sounds instead of random audio segments. It pre-trains this model on roughly 40 hours of public cough recordings and evaluates the resulting representations on five downstream tasks such as COVID-19 detection, COPD state classification, and smoker status prediction. Results indicate that this cough-specific pre-training approach yields higher performance than standard random-cropping techniques when the data consists of cough recordings. The work additionally benchmarks multiple models on COPD classification and reports that even strong pre-trained audio models reach only 57 percent unweighted average recall.

Core claim

CoughPhase-CLR is a contrastive learning method that forms positive pairs by selecting segments from the same acoustic phase within a cough recording. Pre-training with this phase-informed pairing on cough audio produces representations that improve accuracy on multiple respiratory health classification tasks relative to models trained with random cropping or generic audio pre-training.

What carries the argument

CoughPhase-CLR contrastive framework, which constructs positive pairs based on the physiological phases of cough acoustics rather than random temporal crops.

If this is right

  • Cough-specific phase pairing can raise classification accuracy for conditions such as COVID-19 when only cough recordings are available.
  • The method allows more effective use of unlabeled cough data for training diagnostic models.
  • COPD state classification stays difficult, with the best tested models limited to 57 percent UAR.
  • Respiratory-sound pre-training offers an alternative path to speech-based analysis for certain cough-related tasks.

Where Pith is reading between the lines

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

  • Similar phase-aware pairing could be tested on other repetitive physiological sounds such as breathing cycles to improve related models.
  • The technique might reduce the volume of labeled data needed to reach usable accuracy in cough-based health screening.
  • Combining phase-informed cough pre-training with large general audio models remains an open direction that could compound gains.

Load-bearing premise

Constructing positive pairs from segments of the same physiological cough phase produces more useful representations for downstream health classification than random cropping does.

What would settle it

An experiment in which random-cropping pre-training achieves equal or higher accuracy than CoughPhase-CLR across the reported downstream tasks on cough data would falsify the claimed advantage.

Figures

Figures reproduced from arXiv: 2606.21411 by Andreas Triantafyllopoulos, Anton Batliner, Bj\"orn W. Schuller, Marius Moldovan, Thomas M. Berghaus.

Figure 1
Figure 1. Figure 1: Example of two coughs divided into the acoustic cough phases: i) explosive, ii) intermediate, and iii) voiced. Adapted from Eni et al. [7]. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Pre-training framework of CoughPhase-CLR. TABLE I OVERVIEW OF DATASETS USED FOR PRE-TRAINING. (SR: SAMPLING RATE; DURATION: MEAN [95% QUANTILE RANGE]) Dataset Name #Recordings #Coughs Modality SR Duration of coughs (s) UK COVID-19 [15] 19533 47,220 Cough 48 kHz 0.60 [0.43 - 0.96] COUGHVID [14] 7179 85,799 Cough 48 kHz 0.61 [0.42 - 1.02] B. Pre-training Datasets To pre-train CoughPhase-CLR, we used the two … view at source ↗
Figure 3
Figure 3. Figure 3: Comparison of the two pre-training tasks, random crop ( [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Saliency maps illustrating model interpretability on cough spectrograms. The top row displays a cough from the exacerbation class, while the bottom [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Box plots of twelve acoustic features extracted from cough recordings, compared between the stable and exacerbation states. Each box shows the [PITH_FULL_IMAGE:figures/full_fig_p010_5.png] view at source ↗
read the original abstract

In this work, we introduce CoughPhase-CLR, a self-supervised learning framework designed to leverage the physiological phases of a cough for robust representation learning. Unlike generic contrastive frameworks, CoughPhase-CLR constructs positive pairs based on these specific acoustic phases. We pre-trained our model on approximately 40 hours of public cough audio and evaluated it across five downstream tasks, including COVID-19 detection, chronic obstructive pulmonary disease (COPD) state classification, and smoker status prediction. Our results demonstrate that cough-specific pre-training consistently outperforms standard random-cropping techniques when training on cough recordings. Additionally, we benchmarked a diverse set of state-of-the-art models on COPD state classification, highlighting the difficulty of this task. The best-performing models, pretrained on either general audio or respiratory sounds, achieved a UAR of 57\%, failing to outperform the state-of-the-art performance of 84\% UAR achieved using speech analysis.

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

0 major / 3 minor

Summary. The paper introduces CoughPhase-CLR, a self-supervised contrastive learning framework that constructs positive pairs from physiological phases of cough sounds rather than random crops. It pre-trains on ~40 hours of public cough audio and evaluates on five downstream tasks including COVID-19 detection, COPD state classification, and smoker status prediction. The central claim is that phase-informed pre-training consistently outperforms random-cropping baselines; the paper also benchmarks multiple SOTA audio models on COPD classification, reporting a best UAR of 57% that falls short of 84% UAR obtained via speech analysis.

Significance. If the reported gains hold under rigorous validation, the work shows that domain-specific acoustic structure (cough phases) can be leveraged to improve self-supervised representations for respiratory audio, offering a practical route toward specialized foundation models for health diagnostics. The COPD benchmarking usefully documents the difficulty of the task when restricted to cough recordings and may steer future efforts toward multi-modal or speech-augmented approaches. Public-data pre-training supports reproducibility.

minor comments (3)
  1. Abstract: the summary of results would be strengthened by inclusion of at least one or two key quantitative metrics (e.g., UAR deltas or absolute scores) with brief mention of statistical testing.
  2. Abstract / Results: the 84% UAR figure from speech analysis should be accompanied by an explicit citation or reference to the source method.
  3. Methods: provide clearer details on the exact construction of positive pairs from cough phases (e.g., how phase boundaries are detected or annotated) to allow replication.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the positive summary, significance assessment, and recommendation of minor revision. The report correctly captures the core contribution of CoughPhase-CLR and the benchmarking results. No major comments were provided in the report, so we have no points to rebut or revise on that basis. We will address any minor issues identified during the revision process.

Circularity Check

0 steps flagged

No significant circularity; purely empirical claims

full rationale

The manuscript introduces CoughPhase-CLR as a contrastive pre-training framework that constructs positive pairs from cough phases and reports empirical gains over random cropping on five downstream classification tasks. No equations, derivations, fitted parameters renamed as predictions, or uniqueness theorems appear in the provided text. All central claims rest on reported UAR metrics and benchmark comparisons rather than any self-referential reduction. Self-citations, if present, are not load-bearing for any mathematical step. The work is therefore self-contained as an empirical study.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review yields no identifiable free parameters, axioms, or invented entities; all elements are standard in self-supervised audio learning.

pith-pipeline@v0.9.1-grok · 5706 in / 1118 out tokens · 17261 ms · 2026-06-26T12:54:05.970217+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

49 extracted references · 24 canonical work pages · 1 internal anchor

  1. [1]

    This is caused by the sudden release of pressure as the glottis opens [7, 9]

    Theexplosive phaseis characterized by an explosive burst with a very sharp, high-amplitude increase in the sound energy envelope. This is caused by the sudden release of pressure as the glottis opens [7, 9]

  2. [2]

    The energy during this phase is typically lower than the initial burst and gradually decays as airflow decreases [10, 11]

    Theintermediate phasefollows this initial burst and is characterized by a more sustained, high-frequency “noisy” sound generated by the sustained turbulent air- flow through the airways. The energy during this phase is typically lower than the initial burst and gradually decays as airflow decreases [10, 11]

  3. [3]

    the acoustic information in all phases of a cough is equally important for the purposes of TB classification

    Thevoiced phaseis the final phase and is not present in all coughs. It includes a pitch frequency induced by the partial closure of the vocal cords, introducing a periodic, tonal quality to the signal [10]. Coughing sounds are also indicative of an underlying pathology. For instance, coughs from COPD patients typically have a longer duration, a later-occu...

    2021

  4. [4]

    No pre-training eGeMAPS[20]51[50−52] CNN14(scratch) [33]54[52−57] EfficientNet-B0(scratch) [27]51[48−54]

  5. [5]

    Pre-trained on image data EfficientNet-B0(ImageNet) [27]53[50−56] VGG-16-BN[34]51[47−56]

  6. [6]

    Pre-trained on general audio data CNN14(AudioSet) [33]57[52−61] wav2vec2.0[35]50[48−52] HuBERT[36]50[48−51]

  7. [7]

    Pre-trained on respiratory audio HeAR[6]51[50−53] OPERA-CE[2]56[54−57] OPERA-CT[2]57[53−60] OPERA-GT[2]50[47−54]

  8. [8]

    The distributions overlap substantially between the two states, with the majority of features showing no statistically significant difference atα= 0.05

    Pre-trained on cough audio CoughPhase-CLR(ours)53[51−55] these features for the stable and exacerbation states, annotated with p-values from two-sided Mann–Whitney U tests. The distributions overlap substantially between the two states, with the majority of features showing no statistically significant difference atα= 0.05. This suggests that handcrafted ...

    2000

  9. [9]

    Hear4health: A blueprint for making computer audition a staple of modern healthcare,

    A. Triantafyllopoulos, A. Kathan, A. Baird, L. Christ, A. Gebhard, M. Gerczuk, V . Karas, T. H¨ubner, X. Jing, S. Liu, et al., “Hear4health: A blueprint for making computer audition a staple of modern healthcare,” Frontiers in digital health, vol. 5, p. 1 196 079, 2023

    2023

  10. [10]

    Zhang, T

    Y . Zhang, T. Xia, J. Han, Y . Wu, G. Rizos, Y . Liu, M. Mosuily, J. Chauhan, and C. Mascolo,Towards open respiratory acoustic foundation models: Pretraining and benchmarking, 2024.DOI: 10 . 48550/ARXIV.2406.16148 [Online]. Available: https://arxiv.org/abs/ 2406.16148

    arXiv 2024

  11. [11]

    Hjorth Larsen, J

    M. Pahar, M. Klopper, B. Reeve, R. Warren, G. Theron, and T. Niesler, “Automatic cough classification for tuberculosis screening in a real- world environment,”Physiological Measurement, vol. 42, no. 10, p. 105 014, Oct. 2021,ISSN: 1361-6579.DOI: 10.1088/1361- 6579/ ac2fb8 [Online]. Available: http://dx.doi.org/10.1088/1361- 6579/ ac2fb8

    work page doi:10.1088/1361- 2021

  12. [12]

    Wavelet analysis of voluntary cough sound in patients with respiratory diseases,

    J. Knocikova, J. Korpas, M. Vrabec, and M. Javorka, “Wavelet analysis of voluntary cough sound in patients with respiratory diseases,”J Physiol Pharmacol, vol. 59, no. Suppl 6, pp. 331–40, 2008

    2008

  13. [13]

    Cough duration, energy and sound frequency in covid-19 patients: The spectral analysis results,

    A. V . Budnevsky, D. Kosanovic, E. S. Ovsyannikov, O. N. Choporov, A. V . Pertsev, S. N. Feigelman, T. A. Chernik, A. V . Maksimov, G. G. Prozorova, S. A. Kozhevnikova, R. E. Tokmachev, A. V . Belyakova, V . R. Drobysheva, and S. N. Avdeev, “Cough duration, energy and sound frequency in covid-19 patients: The spectral analysis results,” BMC Pulmonary Medi...

    work page doi:10.1186/s12890- 2025

  14. [14]

    S. Baur, Z. Nabulsi, W.-H. Weng, J. Garrison, L. Blankemeier, S. Fishman, C. Chen, S. Kakarmath, M. Maimbolwa, N. Sanjase, B. Shuma, Y . Matias, G. S. Corrado, S. Patel, S. Shetty, S. Prabhakara, M. Muyoyeta, and D. Ardila,Hear – health acoustic representations, 2024.DOI: 10.48550/ARXIV.2403.02522 [Online]. Available: https: //arxiv.org/abs/2403.02522

    work page doi:10.48550/arxiv.2403.02522 2024

  15. [15]

    Cough detection using a non-contact microphone: A nocturnal cough study,

    M. Eni, V . Mordoh, and Y . Zigel, “Cough detection using a non-contact microphone: A nocturnal cough study,”PLOS ONE, vol. 17, no. 1, F. Albu, Ed., e0262240, Jan. 2022,ISSN: 1932-6203.DOI: 10.1371/ journal.pone.0262240 [Online]. Available: http://dx.doi.org/10.1371/ journal.pone.0262240

    2022

  16. [16]

    Acoustic parameters of voluntary cough in healthy non-smoking subjects,

    P. M. Olia, P. Sestini, and M. Vagliasindi, “Acoustic parameters of voluntary cough in healthy non-smoking subjects,”Respirology, vol. 5, no. 3, pp. 271–275, Sep. 2000,ISSN: 1440-1843.DOI: 10.1046/j.1440- 1843.2000.00259.x [Online]. Available: http://dx.doi.org/10.1046/j. 1440-1843.2000.00259.x

    work page doi:10.1046/j.1440- 2000

  17. [17]

    How to quantify coughing: Correlations with quality of life in chronic cough,

    A. Kelsall, S. Decalmer, D. Webster, N. Brown, K. McGuinness, A. Woodcock, and J. Smith, “How to quantify coughing: Correlations with quality of life in chronic cough,”European Respiratory Journal, vol. 32, no. 1, pp. 175–179, Feb. 2008,ISSN: 1399-3003.DOI: 10. 1183/09031936.00101307 [Online]. Available: http://dx.doi.org/10. 1183/09031936.00101307

    arXiv 2008

  18. [18]

    Past and trends in cough sound acquisition, automatic detection and automatic classifi- cation: A comparative review,

    A. Serrurier, C. Neuschaefer-Rube, and R. R ¨ohrig, “Past and trends in cough sound acquisition, automatic detection and automatic classifi- cation: A comparative review,”Sensors, vol. 22, no. 8, p. 2896, Apr. 2022,ISSN: 1424-8220.DOI: 10.3390/s22082896 [Online]. Available: http://dx.doi.org/10.3390/s22082896

    work page doi:10.3390/s22082896 2022

  19. [19]

    The present and future of cough counting tools,

    J. I. Hall, M. Lozano, L. Estrada-Petrocelli, S. Birring, and R. Turner, “The present and future of cough counting tools,”Journal of Thoracic Disease, vol. 12, no. 9, pp. 5207–5223, Sep. 2020,ISSN: 2077-6624. DOI: 10.21037/jtd-2020-icc-003 [Online]. Available: http://dx.doi.org/ 10.21037/jtd-2020-icc-003

    work page doi:10.21037/jtd-2020-icc-003 2020

  20. [20]

    Cough and its importance in copd,

    J. Smith and A. Woodcock, “Cough and its importance in copd,” International Journal of COPD, vol. 1, no. 3, pp. 305–314, Aug. 2006, ISSN: 1176-9106.DOI: 10.2147/copd.2006.1.3.305 [Online]. Available: http://dx.doi.org/10.2147/copd.2006.1.3.305

    work page doi:10.2147/copd.2006.1.3.305 2006

  21. [21]

    On the importance of different cough phases for covid-19 detection,

    Y . Zhu, M. H. Shaik, and T. H. Falk, “On the importance of different cough phases for covid-19 detection,” inICASSP 2023 - 2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), IEEE, Jun. 2023, pp. 1–5.DOI: 10 . 1109 / icassp49357 . 2023 . 10095820 [Online]. Available: http : / / dx . doi . org / 10 . 1109 / ICASSP49357....

    arXiv 2023

  22. [22]

    The coughvid crowdsourcing dataset, a corpus for the study of large-scale cough analysis algo- rithms,

    L. Orlandic, T. Teijeiro, and D. Atienza, “The coughvid crowdsourcing dataset, a corpus for the study of large-scale cough analysis algo- rithms,”Scientific Data, vol. 8, no. 1, Jun. 2021,ISSN: 2052-4463. DOI: 10.1038/s41597-021-00937-4 [Online]. Available: http://dx.doi. org/10.1038/s41597-021-00937-4

    work page doi:10.1038/s41597-021-00937-4 2021

  23. [23]

    Coppock, G

    H. Coppock, G. Nicholson, I. Kiskin, V . Koutra, K. Baker, J. Budd, R. Payne, E. Karoune, D. Hurley, A. Titcomb, S. Egglestone, A. T. Ca˜nadas, L. Butler, R. Jersakova, J. Mellor, S. Patel, T. Thornley, P. Diggle, S. Richardson, J. Packham, B. W. Schuller, D. Pigoli, S. Gilmour, S. Roberts, and C. Holmes,Audio-based ai classifiers show no evidence of impr...

    arXiv 2022

  24. [24]

    Coswara: A respiratory sounds and symptoms dataset for remote screening of sars-cov-2 infection,

    D. Bhattacharya, N. K. Sharma, D. Dutta, S. R. Chetupalli, P. Mote, S. Ganapathy, C. Chandrakiran, S. Nori, K. K. Suhail, S. Gonuguntla, and M. Alagesan, “Coswara: A respiratory sounds and symptoms dataset for remote screening of sars-cov-2 infection,”Scientific Data, vol. 10, no. 1, Jun. 2023,ISSN: 2052-4463.DOI: 10.1038/s41597-023-02266-0 [Online]. Avai...

    work page doi:10.1038/s41597-023-02266-0 2023

  25. [25]

    Covid-19 detection in cough, breath and speech using deep transfer learning and bottleneck features,

    M. Pahar, M. Klopper, R. Warren, and T. Niesler, “Covid-19 detection in cough, breath and speech using deep transfer learning and bottleneck features,”Computers in Biology and Medicine, vol. 141, p. 105 153, Feb. 2022,ISSN: 0010-4825.DOI: 10.1016/j.compbiomed.2021.105153 [Online]. Available: http://dx.doi.org/10.1016/j.compbiomed.2021. 105153

    work page doi:10.1016/j.compbiomed.2021.105153 2022

  26. [26]

    Advancing cough classification: Swin transformer vs. 2d cnn with stft and augmentation techniques,

    M. Ghourabi, F. Mourad-Chehade, and A. Chkeir, “Advancing cough classification: Swin transformer vs. 2d cnn with stft and augmentation techniques,”Electronics, vol. 13, no. 7, p. 1177, Mar. 2024,ISSN: 2079-9292.DOI: 10 . 3390 / electronics13071177 [Online]. Available: http://dx.doi.org/10.3390/electronics13071177

    work page doi:10.3390/electronics13071177 2024

  27. [27]

    Exploring automatic diagnosis of covid-19 from crowdsourced respiratory sound data,

    C. Brown, J. Chauhan, A. Grammenos, J. Han, A. Hasthanasombat, D. Spathis, T. Xia, P. Cicuta, and C. Mascolo, “Exploring automatic diagnosis of covid-19 from crowdsourced respiratory sound data,” 2020.DOI: 10.48550/ARXIV.2006.05919 [Online]. Available: https: //arxiv.org/abs/2006.05919

    work page doi:10.48550/arxiv.2006.05919 2020

  28. [28]

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

    F. Eyben, K. R. Scherer, B. W. Schuller, J. Sundberg, E. Andre, C. Busso, L. Y . Devillers, J. Epps, P. Laukka, S. S. Narayanan, and K. P. Truong,The geneva minimalistic acoustic parameter set (gemaps) for voice research and affective computing, Apr. 2016.DOI: 10.1109/taffc. 2015.2457417 [Online]. Available: http://dx.doi.org/10.1109/TAFFC. 2015.2457417

    work page doi:10.1109/taffc 2016

  29. [29]

    CNN architectures for large-scale audio clas- sification,

    S. Hershey, S. Chaudhuri, D. P. W. Ellis, J. F. Gemmeke, A. Jansen, R. C. Moore, M. Plakal, D. Platt, R. A. Saurous, B. Seybold, M. Slaney, R. J. Weiss, and K. Wilson, “Cnn architectures for large-scale audio classification,” in2017 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), IEEE, Mar. 2017, pp. 131– 135.DOI: 10 . 1...

    work page doi:10.1109/icassp.2017.7952132 2017

  30. [30]

    In: NeurIPS (2022)

    P.-Y . Huang, H. Xu, J. Li, A. Baevski, M. Auli, W. Galuba, F. Metze, and C. Feichtenhofer,Masked autoencoders that listen, 2022.DOI: 10.48550/ARXIV.2207.06405 [Online]. Available: https://arxiv.org/ abs/2207.06405

    work page doi:10.48550/arxiv.2207.06405 2022

  31. [31]

    Y . Wu, K. Chen, T. Zhang, Y . Hui, M. Nezhurina, T. Berg-Kirkpatrick, and S. Dubnov,Large-scale contrastive language-audio pretraining with feature fusion and keyword-to-caption augmentation, 2022.DOI: 10.48550/ARXIV.2211.06687 [Online]. Available: https://arxiv.org/ abs/2211.06687

    work page doi:10.48550/arxiv.2211.06687 2022

  32. [32]

    Niizumi, D

    D. Niizumi, D. Takeuchi, M. Yasuda, B. T. Nguyen, Y . Ohishi, and N. Harada,Towards pre-training an effective respiratory audio foundation model, 2025.DOI: 10.48550/ARXIV.2505.15307 [Online]. Available: https://arxiv.org/abs/2505.15307

    work page doi:10.48550/arxiv.2505.15307 2025

  33. [33]

    Q. Wang, Z. Bu, J. Mao, W. Zhu, J. Zhao, W. Du, G. Shi, M. Zhou, S. Chen, and J. Qu,Towards reliable respiratory disease diagnosis based on cough sounds and vision transformers, 2024.DOI: 10.48550/ ARXIV.2408.15667 [Online]. Available: https://arxiv.org/abs/2408. 15667

    arXiv 2024

  34. [34]

    Foret, A

    P. Foret, A. Kleiner, H. Mobahi, and B. Neyshabur,Sharpness-aware minimization for efficiently improving generalization, 2020.DOI: 10. 48550/ARXIV.2010.01412 [Online]. Available: https://arxiv.org/abs/ 2010.01412

    Pith/arXiv arXiv 2020

  35. [35]

    EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks

    M. Tan and Q. V . Le, “Efficientnet: Rethinking model scaling for convolutional neural networks,” 2019.DOI: 10.48550/ARXIV.1905. 11946 [Online]. Available: https://arxiv.org/abs/1905.11946

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1905 2019

  36. [36]

    T. Chen, S. Kornblith, M. Norouzi, and G. Hinton,A simple framework for contrastive learning of visual representations, 2020.DOI: 10.48550/ ARXIV.2002.05709 [Online]. Available: https://arxiv.org/abs/2002. 05709

    Pith/arXiv arXiv 2020

  37. [37]

    Sustained vowels for pre- vs post-treatment copd classification,

    A. Triantafyllopoulos, A. Batliner, W. Mayr, M. Fendler, F. Pokorny, M. Gerczuk, S. Amiriparian, T. Berghaus, and B. Schuller, “Sustained vowels for pre- vs post-treatment copd classification,” inInterspeech 2024, ISCA, 2024.DOI: 10 . 21437 / Interspeech . 2024 - 96 [Online]. Available: https://arxiv.org/abs/2406.06355

    arXiv 2024

  38. [38]

    Assessing the clinical and functional status of copd patients using speech analysis during and after exacerbation,

    W. Mayr, A. Triantafyllopoulos, A. Batliner, B. Schuller, and T. Berghaus, “Assessing the clinical and functional status of copd patients using speech analysis during and after exacerbation,”International Journal of Chronic Obstructive Pulmonary Disease, vol. V olume 20, pp. 137–147, Jan. 2025,ISSN: 1178-2005.DOI: 10.2147/copd.s480842 [Online]. Available:...

    work page doi:10.2147/copd.s480842 2025

  39. [39]

    Usefulness of the modified 0-10 borg scale in assessing the degree of dyspnea in patients with copd and asthma,

    K. R. Kendrick, S. C. Baxi, and R. M. Smith, “Usefulness of the modified 0-10 borg scale in assessing the degree of dyspnea in patients with copd and asthma,”Journal of Emergency Nursing, vol. 26, no. 3, pp. 216–222, Jun. 2000,ISSN: 0099-1767.DOI: 10 . 1016 / s0099 - 1767(00)90093-x [Online]. Available: http://dx.doi.org/10.1016/s0099- 1767(00)90093-x

    work page doi:10.1016/s0099- 2000

  40. [40]

    The copd assessment test: A systematic review,

    N. Gupta, L. M. Pinto, A. Morogan, and J. Bourbeau, “The copd assessment test: A systematic review,”European Respiratory Journal, vol. 44, no. 4, pp. 873–884, Jul. 2014,ISSN: 1399-3003.DOI: 10.1183/ 09031936.00025214 [Online]. Available: http://dx.doi.org/10.1183/ 09031936.00025214

    arXiv 2014

  41. [41]

    Q. Kong, Y . Cao, T. Iqbal, Y . Wang, W. Wang, and M. D. Plumbley, Panns: Large-scale pretrained audio neural networks for audio pattern recognition, 2019.DOI: 10 . 48550 / ARXIV. 1912 . 10211 [Online]. Available: https://arxiv.org/abs/1912.10211

    arXiv 2019

  42. [42]

    Very deep convolutional networks for large-scale image recognition,

    K. Simonyan and A. Zisserman, “Very deep convolutional networks for large-scale image recognition,”arXiv preprint arXiv:1409.1556, 2014

    Pith/arXiv arXiv 2014

  43. [43]

    Baevski, Y

    A. Baevski, H. Zhou, A. Mohamed, and M. Auli,Wav2vec 2.0: A framework for self-supervised learning of speech representations, 2020.DOI: 10.48550/ARXIV.2006.11477 [Online]. Available: https: //arxiv.org/abs/2006.11477

    work page doi:10.48550/arxiv.2006.11477 2020

  44. [44]

    IEEE/ACM Trans

    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,ISSN: 2329-9304.DOI: 10.1109/taslp.2021.3122291 [Online]. Available: http://dx.doi.or...

    work page doi:10.1109/taslp.2021.3122291 2021

  45. [45]

    Specaugment: A simple data augmentation method for automatic speech recognition,

    D. S. Park, W. Chan, Y . Zhang, C.-C. Chiu, B. Zoph, E. D. Cubuk, and Q. V . Le, “Specaugment: A simple data augmentation method for automatic speech recognition,” 2019.DOI: 10.48550/ARXIV.1904. 08779 [Online]. Available: https://arxiv.org/abs/1904.08779

    work page doi:10.48550/arxiv.1904 2019

  46. [46]

    Distinguishing between pre- and post-treatment in the speech of patients with chronic obstructive pulmonary disease,

    A. Triantafyllopoulos, M. Fendler, A. Batliner, M. Gerczuk, S. Amiri- parian, T. Berghaus, and B. W. Schuller, “Distinguishing between pre- and post-treatment in the speech of patients with chronic obstructive pulmonary disease,” inInterspeech 2022, ISCA, Sep. 2022, pp. 3623– 3627.DOI: 10 . 21437 / interspeech . 2022 - 10333 [Online]. Available: http://dx...

    work page doi:10.21437/interspeech.2022-10333 2022

  47. [47]

    Y . Tian, C. Sun, B. Poole, D. Krishnan, C. Schmid, and P. Isola,What makes for good views for contrastive learning?2020.DOI: 10.48550/ ARXIV.2005.10243 [Online]. Available: https://arxiv.org/abs/2005. 10243

    arXiv 2020

  48. [48]

    Charting 15 years of progress in deep learning for speech emotion recognition: A replication study,

    A. Triantafyllopoulos, A. Batliner, and B. W. Schuller, “Charting 15 years of progress in deep learning for speech emotion recognition: A replication study,”IEEE Transactions on Affective Computing, 2026

    2026

  49. [49]

    Detecting copd through speech analysis: A dataset of danish speech and machine learning approach,

    C. Sankey-Olsen, R. H. Olesen, T. O. Eberhard, A. Triantafyllopoulos, B. Schuller, and I. Aslan, “Detecting copd through speech analysis: A dataset of danish speech and machine learning approach,”arXiv preprint arXiv:2508.02354, 2025

    arXiv 2025