pith. sign in

arxiv: 2410.03280 · v1 · pith:O4WTDFH6new · submitted 2024-10-04 · 📡 eess.AS · cs.AI· cs.LG· eess.SP

Manikin-Recorded Cardiopulmonary Sounds Dataset Using Digital Stethoscope

Yasaman Torabi , Shahram Shirani , James P. Reilly This is my paper

Pith reviewed 2026-05-23 19:51 UTC · model grok-4.3

classification 📡 eess.AS cs.AIcs.LGeess.SP
keywords cardiopulmonary soundsdigital stethoscopemanikin datasetheart soundslung soundsaudio datasetdisease detectionsound separation
0
0 comments X

The pith

A new dataset records both separate and mixed heart and lung sounds from a clinical manikin using a digital stethoscope.

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

The paper presents a collection of audio recordings captured from a manikin simulator designed to replicate human physiology. It supplies both individual heart and lung sounds as well as their combinations at multiple chest locations, covering normal cases and a range of abnormalities including murmurs, atrial fibrillation, wheezing, crackles, and others. The authors state this is the first dataset to provide both separate and mixed cardiorespiratory sounds. The recordings are intended to support artificial intelligence work on automated disease detection, sound classification, separation of mixed audio, and related deep learning tasks in audio signal processing.

Core claim

The authors assembled a dataset of digital stethoscope recordings taken from a clinical manikin at anatomical sites chosen by specialist nurses; each recording contains either isolated heart sounds, isolated lung sounds, or their mixtures, with both normal physiology and listed pathologies present, and frequency filters applied to emphasize particular sound components.

What carries the argument

The clinical manikin as the controlled source of clean cardiopulmonary sounds recorded at different body locations.

If this is right

  • The dataset supplies labeled examples for supervised classification of specific abnormalities such as murmurs or crackles.
  • It supplies paired separate and mixed recordings that can be used to develop and test unsupervised audio separation methods.
  • It provides training material for deep learning models aimed at cardiopulmonary sound analysis without requiring access to patient data.
  • Recordings at multiple anatomical locations allow models to learn location-specific sound patterns.

Where Pith is reading between the lines

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

  • The clean manikin environment could serve as a controlled starting point for separation algorithms before they are tested on noisier clinical recordings.
  • If models trained here generalize poorly to real patients, the dataset would still be useful for rapid prototyping of algorithms that later require fine-tuning on real data.
  • The same manikin-based collection approach could be extended to create parallel datasets for other internal sounds such as bowel or joint audio.

Load-bearing premise

The sounds generated by the clinical manikin sufficiently mimic real human cardiopulmonary sounds to be useful for training AI models that will be applied to actual patients.

What would settle it

Train a disease-detection model on the manikin recordings and measure its accuracy on a held-out set of real patient recordings; a substantial performance gap relative to models trained directly on real data would show the dataset does not transfer.

Figures

Figures reproduced from arXiv: 2410.03280 by James P. Reilly, Shahram Shirani, Yasaman Torabi.

Figure 1
Figure 1. Figure 1: FIGURE 1 [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIGURE 3 [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 6
Figure 6. Figure 6: FIGURE 6 [PITH_FULL_IMAGE:figures/full_fig_p004_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: FIGURE 7 [PITH_FULL_IMAGE:figures/full_fig_p005_7.png] view at source ↗
read the original abstract

Heart and lung sounds are crucial for healthcare monitoring. Recent improvements in stethoscope technology have made it possible to capture patient sounds with enhanced precision. In this dataset, we used a digital stethoscope to capture both heart and lung sounds, including individual and mixed recordings. To our knowledge, this is the first dataset to offer both separate and mixed cardiorespiratory sounds. The recordings were collected from a clinical manikin, a patient simulator designed to replicate human physiological conditions, generating clean heart and lung sounds at different body locations. This dataset includes both normal sounds and various abnormalities (i.e., murmur, atrial fibrillation, tachycardia, atrioventricular block, third and fourth heart sound, wheezing, crackles, rhonchi, pleural rub, and gurgling sounds). The dataset includes audio recordings of chest examinations performed at different anatomical locations, as determined by specialist nurses. Each recording has been enhanced using frequency filters to highlight specific sound types. This dataset is useful for applications in artificial intelligence, such as automated cardiopulmonary disease detection, sound classification, unsupervised separation techniques, and deep learning algorithms related to audio signal processing.

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

1 major / 0 minor

Summary. The paper presents a new publicly released dataset of cardiopulmonary sounds (heart and lung, normal and abnormal) recorded from a clinical manikin using a digital stethoscope. It includes both separate and mixed cardiorespiratory sounds captured at multiple anatomical locations, with the claim that this is the first such dataset offering both separate and mixed recordings. The recordings are described as enhanced via frequency filters, and the dataset is positioned for downstream AI tasks including classification, separation, and disease detection.

Significance. If the dataset is made available as described and the manikin sounds prove representative, it would provide a clean, labeled resource for audio-signal-processing and machine-learning research on cardiopulmonary sounds, particularly for tasks requiring mixed vs. separated sources. The inclusion of multiple abnormality types at controlled locations is a practical strength for supervised learning benchmarks.

major comments (1)
  1. [Abstract] Abstract: the statement that the dataset 'is useful for applications in artificial intelligence, such as automated cardiopulmonary disease detection' on real patients rests on the unverified assumption that manikin-generated sounds sufficiently replicate human physiology; no validation metrics, spectral comparisons, or error analysis against real-patient recordings are provided to support transferability.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the detailed review and constructive feedback. We address the single major comment below and agree that a revision to the abstract is warranted to avoid overstating transferability.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the statement that the dataset 'is useful for applications in artificial intelligence, such as automated cardiopulmonary disease detection' on real patients rests on the unverified assumption that manikin-generated sounds sufficiently replicate human physiology; no validation metrics, spectral comparisons, or error analysis against real-patient recordings are provided to support transferability.

    Authors: We agree that the abstract's phrasing implies prospective utility for real-patient disease detection without providing supporting evidence of acoustic similarity between the manikin recordings and human physiology. The manuscript positions the dataset as a controlled, labeled benchmark for tasks such as classification and source separation, where the manikin provides repeatable normal and pathological variants at known locations. No spectral comparisons or validation against real-patient recordings are included, as the work focuses on dataset creation rather than clinical equivalence. We will revise the abstract to qualify the AI applications, removing the direct reference to automated cardiopulmonary disease detection on real patients and instead highlighting utility for benchmark development and method prototyping. revision: yes

Circularity Check

0 steps flagged

No circularity; purely descriptive dataset release

full rationale

The paper contains no equations, derivations, fitted parameters, predictions, or mathematical claims of any kind. Its central contribution is the public release of audio recordings captured from a clinical manikin, with novelty stated only as a standard 'to our knowledge' qualifier. No load-bearing step reduces to a self-citation, self-definition, or fitted input renamed as output. The usefulness statements for downstream AI are explicitly caveated by the manikin source and do not constitute a derivation chain.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

This paper introduces no free parameters, axioms, or invented entities as it is a description of an empirical dataset rather than a theoretical or modeling contribution.

pith-pipeline@v0.9.0 · 5738 in / 1225 out tokens · 38729 ms · 2026-05-23T19:51:56.009037+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

18 extracted references · 18 canonical work pages · 2 internal anchors

  1. [1]

    CirCor DigiScope Phonocardiogram

    Dataset Paper Copyright: © 2024 by the authors. Submitted under the terms and conditions of the Creative Commons Attribution (CC BY) license 1 Manikin-Recorded Cardiopulmonary Sounds Dataset Using Digital Stethoscope Yasaman Torabi1, Shahram Shirani1,2, James P. Reilly1 1Electrical and Computer Engineering Department, McMaster University, Hamilton, Ontari...

    work page doi:10.17632/8972jxbpmp.1 2024

  2. [2]

    [14], who focused solely on lung sounds and heart sounds, respectively, our dataset includes heart, lung, and mixed recordings

    and Oliveira et al. [14], who focused solely on lung sounds and heart sounds, respectively, our dataset includes heart, lung, and mixed recordings. In 2022, Julio Alejandro Valdez et al. gathered a cardiopulmonary dataset which provided separate heart and lung sound files recorded simultaneously but did not offer mixed recordings [15]. Unlike this work, w...

    work page 2022

  3. [3]

    as follows: Dataset Paper Copyright: © 2024 by the authors. Submitted under the terms and conditions of the Creative Commons Attribution (CC BY) license 3 - Apex (A): Mitral area - Right Upper Sternal Border (RUSB): Aortic area - Left Upper Sternal Border (LUSB): Pulmonary area - Left Lower Sternal Border (LLSB): Tricuspid area - Right Costal Margin (RC) ...

    work page 2024

  4. [4]

    b Rapid decay @10 dB/octave below 200 Hz

    Technical Specification of the Digital Stethoscope Parameter Value Frequency Response (Bell Mode) [20-200] Hz a Frequency Response (Diaphragm Mode) [100-500] Hz b Frequency Response (Midrange Mode) [50-500] Hz c Amplification Up to 40x Recording Sample Rate 22050 Hz Active Noise Cancellation 85% d a Rapid decay @20 dB/octave above 300 Hz. b Rapid decay @1...

    work page 2024

  5. [5]

    Design and Implementation of an Ultralow-Power ECG Patch and Smart Cloud-Based Platform,

    B. Baraeinejad et al., "Design and Implementation of an Ultralow-Power ECG Patch and Smart Cloud-Based Platform," IEEE Transactions on Instrumentation and Measurement, vol. 71, pp. 1-11, 2022, Art no. 2506811, doi: 10.1109/TIM.2022.3164151

    work page doi:10.1109/tim.2022.3164151 2022

  6. [6]

    Auscultation of the respiratory system,

    M. Sarkar, I. Madabhavi, N. Niranjan, and M. Dogra, "Auscultation of the respiratory system," Ann Thorac Med., vol. 10, no. 3, pp. 158–168, Jul.-Sep. 2015, doi: 10.4103/1817-1737.160831

    work page doi:10.4103/1817-1737.160831 2015

  7. [7]

    Evolution of the stethoscope,

    P. J. Bishop, "Evolution of the stethoscope," J. R. Soc. Med., vol. 73, pp. 448–456, 1980, doi: 10.1177/014107688007300611

    work page doi:10.1177/014107688007300611 1980

  8. [8]

    Auscultation in Flight: Comparison of Conventional and Electronic Stethoscopes,

    J. P. Tourtier et al., "Auscultation in Flight: Comparison of Conventional and Electronic Stethoscopes," Air Med. J., vol. 30, pp. 158–160, 2011, doi: 10.1016/j.amj.2010.11.009

    work page doi:10.1016/j.amj.2010.11.009 2011

  9. [10]

    Exploring Sensing Devices for Heart and Lung Sound Monitoring,

    Y. Torabi et al., "Exploring Sensing Devices for Heart and Lung Sound Monitoring," arXiv preprint, doi: 2406.12432,

    work page arXiv

  10. [11]

    Review on the advancements of stethoscope types in chest auscultation,

    J. J. Seah et al., "Review on the advancements of stethoscope types in chest auscultation," Diagnostics, vol. 13, no. 9, p. 1545, 2023, doi: 10.3390/diagnostics13091545

    work page doi:10.3390/diagnostics13091545 2023

  11. [12]

    Clinical IoT in Practice: A Novel Design and Implementation of a Multi-functional Digital Stethoscope for Remote Health Monitoring,

    B. Baraeinejad et al., "Clinical IoT in Practice: A Novel Design and Implementation of a Multi-functional Digital Stethoscope for Remote Health Monitoring," TechRxiv Preprints, Nov. 7, 2023, doi: 10.36227/techrxiv.24459988.v1

    work page doi:10.36227/techrxiv.24459988.v1 2023

  12. [13]

    A Review of Automatic Cardiac Segmentation using Deep Learning and Deformable Models,

    B. Rahmatikaregar et al., "A Review of Automatic Cardiac Segmentation using Deep Learning and Deformable Models," in Artificial Intelligence in Healthcare and Medicine, 1st ed., CRC Press, 2022, pp. 54, doi: 10.1201/9781003120902

    work page doi:10.1201/9781003120902 2022

  13. [14]

    Blind Monaural Source Separation on Heart and Lung Sounds Based on Periodic-Coded Deep Autoencoder,

    K.-H. Tsai et al., "Blind Monaural Source Separation on Heart and Lung Sounds Based on Periodic-Coded Deep Autoencoder," IEEE J. Biomed. Health Inform., vol. 24, no. 11, pp. 3203-3214, 2020, doi: 10.1109/JBHI.2020.3016831

    work page doi:10.1109/jbhi.2020.3016831 2020

  14. [15]

    A dataset of lung sounds recorded from the chest wall using an electronic stethoscope,

    L. Fraiwan et al., "A dataset of lung sounds recorded from the chest wall using an electronic stethoscope," Data in Brief, vol. 34, 2021, doi: 10.1016/j.dib.2021.106000

    work page doi:10.1016/j.dib.2021.106000 2021

  15. [16]

    Cardiopulmonary sounds database,

    J. A. Valdez and P. Mayorga, "Cardiopulmonary sounds database," IEEE Dataport, 2022, doi: 10.21227/8jzz-3x76

    work page doi:10.21227/8jzz-3x76 2022

  16. [17]

    A New Non-Negative Matrix Factorization Approach for Blind Source Separation of Cardiovascular and Respiratory Sound Based on the Periodicity of Heart and Lung Function,

    Y. Torabi et al., "A New Non-Negative Matrix Factorization Approach for Blind Source Separation of Cardiovascular and Respiratory Sound Based on the Periodicity of Heart and Lung Function," arXiv preprint, arXiv:2305.01889,

    work page internal anchor Pith review arXiv

  17. [18]

    Dirac operator on spinors and diffeomorphisms

    P. S. Rao, "Diagnosis of cardiac murmurs in children," Vessel Plus, vol. 6, pp. 22, 2022, doi: 10.20517/2574-1209.2021.105

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.20517/2574-1209.2021.105 2022

  18. [19]

    On potential effectiveness of integration of 3M Littmann 3200 electronic stethoscopes into the third-party diagnostic systems with auscultation signal processing,

    V. Oliynik, "On potential effectiveness of integration of 3M Littmann 3200 electronic stethoscopes into the third-party diagnostic systems with auscultation signal processing," in Proc. 2015 IEEE 35th Int. Conf. Electronics and Nanotechnology (ELNANO), Kyiv, Ukraine, 2015, pp. 417–421, doi: 10.1109/ELNANO.2015.7146923

    work page doi:10.1109/elnano.2015.7146923 2015