Respiratory Disease Prediction Using Respiratory

Sound
Mr. Rajesh T H Kiran Nagayya Hiremath Manoj N R
Dept.of.CS&E Dept.of.CS&E Dept.of.CS&E
PESITM PESITM PESITM
Shimoga, India Shimoga, India Shimoga, India
rajesh@pestrust.edu.in kirannh708@gmail.com manojneswi.4@gmail.com

Shashank P B Suresh B
Dept.of.CS&E Dept.of.CS&E
PESITM PESITM
Shimoga, India Shimoga, India
shashankc1432@gmail.com sureshraghu887@gmail.com

Abstract - This project presents a novel approach to offering a convenient and accessible tool for early detection and
predicting respiratory diseases using voice analysis and machine monitoring of respiratory health.
learning. By leveraging the nuances in vocal characteristics,
such as pitch, harmonics, and spectral features, the system aims The integration of voice analysis with machine learning not only
to identify conditions like Chronic Obstructive Pulmonary provides a cost effective and scalable solution but also enhances
Disease (COPD] asthma, and pneumonia from voice recordings. patient care by enabling continuous monitoring and timely
The methodology involves collecting voice data، preprocessing it interventions. As research in this field progresses, the potential for
to enhance quality, and extracting relevant features using widespread adoption of voice-based diagnostic tools grows,
techniques like Mel-spectrograms. These features are then used marking a significant step forward in respiratory health
to train a Random Forest classifier, a robust machine learning management.
model known for its accuracy and reliability. The trained model
is capable of analyzing new voice inputs to predict respiratory
diseases.
II. RELATED WORK
Keywords – Sound Analysis, Respiratory disease prediction,
Random Forest Algorithm, Feature Extraction, Non-Invasive Respiratory well-grounded psychoanalysis has emerged as a polar
Diagnostics. realm in healthcare, enabling crude detecting and diagnosing of
respiratory diseases. By leveraging car learning techniques, systems
I. Introduction have been improved to take apart respiratory sounds، addressing
challenges like data variableness and real—time pertinency.
respiratory diseases such as Chronic Obstructive Pulmonary Disease
(COPD]، asthma، and pneumonia، represent significant global M. Jannach et al. projected a crossed arrangement combining
health challenges، affecting millions of individuals and placing a sport origin techniques like Mel frequence Cepstral Coefficients
heavy burden on healthcare system. Early detection and continuous (MFCC] with sorting models for respiratory disease foretelling. Thi
monitoring are crucial for managing these conditions effectively an s glide path graveled disease detective work truth using unhurried—
d improving patient outcomes. Traditionally، diagnosing respiratory particularized complete patterns. likewise، A. R. metalworker et al.
diseases involves clinical tests and imaging which can be time used sound data to key out anomalies like wheezing and crackles fo
consuming, costly, and inaccessible to many patients, particularly r diagnosing conditions such as asthma attack and inveterate
remote areas. preventative pulmonic disease (COPD], highlighting the role of
talking supported features in medical exam predictions.
Advancements in machine learning and voice analysis have paved t
he way for innovative، noninvasive diagnostic methods. Voice See through Vector Machines [SVMs) have well—tried existent
analysis in particular, has emerged as a promising tool for predicting in respiratory reasonable sorting. T.D.N.L.H Van et al. incontestable
respiratory diseases. The human voice carries valuable information that SVMs might separate respiratory conditions expeditiously even
about the state of the respiratory system and subtle changes in voice with controlled datasets. S. K. Sharma et al. swollen the feeler by
characteristics can indicate underlying health issues. By analyzing incorporating three d depth psychology, such as combining
features such as pitch harmonics، and spectral components of voice respiratory phone data with environmental factors to amend
recordings it is possible to detect patterns that correlate with diagnosing truth.
respiratory conditions.
A refreshing feeler by H. Lee et al. used real—time data from
This project aims to leverage these advancements by developing a wearable devices to admonisher respiratory sounds، applying deep
system that uses voice analysis to predict respiratory diseases. The learning techniques like Convolutional neuronic Networks (CNNs)
process involves collecting voice recordings from individuals، for around—the clock health tracking. spell this coming increased
preprocessing the audio to enhance quality, extracting relevant real time characteristic capabilities it also stressed the grandness of
features, and training a machine learning model to recognize handling secrecy concerns and ensuring abidance with healthcare
patterns indicative of respiratory diseases. The trained model can regulations such as HIPAA and GDPR.
then analyze new voice inputs to provide accurate predictions,

XXX-X-XXXX-XXXX-X/XX/$XX.00 ©20XX IEEE

Explainability corpse a vituperative dispute in respiratory profound Characteristic origin is centered to respiratory audio depth
psychoanalysis. G. M. H. Jung et al. heavy the need for gauzy psychology. Techniques such as Mel-often Ness Cepstral
models to construct trustingness in systems diagnosing conditions Coefficients [MFCC] and choleric Time Fourier transmute [STFT)
founded on respiratory audio frequency, where decisions now have been wide used. Zhang et al. [7] planned a crossed role model
encroachment patients' lives. some other government issue is the combining MFCC and spectrograph features to raise the
"cold set out job," where controlled first data reduces unit carrying categorization of wheezing and crackling sounds. Their draw near
out. loan blend models that fuse tripartitions sign processing with efficaciously self-addressed data sparseness and variableness,
latest motorcar learning techniques have shown foretell in ensuring buirdly functioning intersectant different datasets.
overcoming these challenges، as pictorial by L. C. Tan et al.
Crossed models which trust denary motorcar learning
These advancements show the growing latent of respiratory techniques have incontestable important anticipate in respiratory
profound analytic thinking systems to transmute healthcare by righteous depth psychology. Bhat and Aishwarya [8] introduced a
enabling non invading real time, and veracious diagnosing of intercrossed poser integrating convolutional nervous networks
respiratory diseases. [CNNs] with continual nervous networks [RNNs]. This unit
processes spectrograms for special features and uses earthly models
for analyzing the sequent natural state of respiratory cycles
improving truth in disease foresight. publicly forthcoming datasets
III. LITRATURE REVIEW: have catalyzed enquiry in respiratory intelligent depth psychology.
The ICBHI 2017 database [9] containing tagged respiratory sounds
Respiratory voice psychoanalysis plays a polar role in old has been instrumental case in training and evaluating political
disease foresight and healthcare conception, leveraging machine learning models. Researchers have used this dataset to
advancements in point processing and auto learning techniques. bench mark algorithms and corroborate their generalizability
These systems aim to take apart non-intrusive sound data, enabling crosswise versatile no subjective scenarios.
exact diagnosing of respiratory conditions like bronchial asthma
lobar pneumonia and habitual preventative pneumonic disease Real-time monitoring systems using wearable devices have also
[COPD). This department reviews existing methodologies datasets, emerged. Lee et al. [10] given a wearable supported respiratory
and car learning approaches in respiratory reasonable depth monitoring unit employing deep learning models for detecting
psychology. respiratory distraint. This attack enhances archaean spotting
capabilities specially for addicted respiratory conditions.
Respiratory reasonable supported symptomatic systems have
garnered tending for their power to discover anomalies like Concealment and restrictive challenges are pivotal in respiratory
wheezing and crackles in sound signals. Isinkaye et al. [1] stressed t secure systems. Jung et al. [11] distressed the need for crystalline a
he grandness of sport origin techniques in healthcare systems، nd explainable models to ascertain user combine. Complying with
highlighting how reasonable processing forms the instauration for healthcare regulations like HIPAA and GDPR، they stressed
disease anticipation algorithms. These insights are discriminative fo sheltered handling of photosensitive respiratory well-grounded data.
r designing prognostic models that can embrace to different
respiratory datasets. Cold head start problems in respiratory systems where small—scale
first data reduces forecasting truth, rest a gainsay. intercrossed
motorcar learning and big data techniques have importantly recommender systems such as the one projected by Tan et al. [12]
increased respiratory disease foretelling. Banu and Gomathy [2] use collaborative and capacity-founded filtering to overthrown data
projected a reasonable categorization organization using data limitations, ensuring homogeneous characteristic functioning even
mining techniques for developed characteristic truth. Their approach with minimum training data.
shot underscored the grandness of preprocessing audio frequency
signals to cover dissonance and variableness, which are built-in Advances in explicable AI (XAI) have promote increased the
challenges in respiratory auditory sensation datasets. Wang et al. [3] interpretability of respiratory voice systems. Patel et al. [13] mature
mature a intercrossed simulation combining sound have abstraction an XAI model for analyzing respiratory sound, allowing clinicians
with deep learning techniques for close foretelling of respiratory to fancy the share of particularized features like wheezing or
anomalies. By integrating signalize processing with simple machine crackles to the final exam anticipation.
learning, they incontestable the feasibleness of non-invading
symptomatic systems. In finale respiratory voice analytic thinking is a quickly evolving
battleground offering vast expected for non invading، straight, and
misbegotten linguistic communication Processing [NLP] ascendable symptomatic of solutions. The consolidation of
techniques have also been modified to respiratory speech sound sophisticated boast abstraction, vigorous car learning models, and
categorization to heighten trial explainability. Asif et al. [4] real—time monitoring capabilities continues to thrust innovations i
introduced a bi directing sound processing framing that combines n healthcare, paving the way for personal and approachable
words features with respiratory talking patterns for graveled respiratory disease foresight systems....
predictions. ThIs approaching illustrates the versatility of
respiratory fathom systems which are now open to of handling
diligent interactions in real—time environments.
IV. METHODOLOGY
Gupta et al. [5] enforced an auto learning supported answer that
uses respiratory audio frequency signals to forebode conditions such 1. Data Gathering: Data gathering is the backbone of building a
as bronchitis and pulmonary emphysema. The arrangement strong voice based diagnostic tool for respiratory disease prediction.
integrates historic respiratory data to hand over individualized This involves very careful and high-quality gathering of voice
characteristic insights, allowing healthcare professionals to make recordings from a varied population of participants, which may
prove supported decisions. likewise، Chen et al. [6] explored defile include those with respiratory diseases such as asthma, COPD, and
supported systems to examine respiratory data in real—time bronchitis and healthy individuals.
improving the scalability and availableness of characteristic
solutions.
2. Preprocessing: Preprocessing is an important step in the Fig 1: Use Case diagram
methodology for predicting respiratory diseases through voice
analysis: it ensures that the quality and consistency of the audio data
used in the analysis. It is a process that starts from noise reduction
where background noises and echoes are filtered out in order to
make the voice clear. Volume normalization adjusts recordings to a
constant range, and equalization balances the frequency spectrum
for enhanced audio quality.

3. Feature Extraction: Feature extraction is the most important step

in the process of predicting respiratory diseases using voice analysis,
as it deals with the identification and isolation of relevant
characteristics from voice recordings that can be used by machine
learning models. This step begins with extracting Mel-frequency
cepstral coefficients (MFCCs), which capture the short-term power
spectrum of the voice and mimic the human ear's perception of
sound, making them particularly useful for distinguishing different
phonemes.

4. Model Training: The process begins with preparing the dataset,

where collected and preprocessed voice data, along with extracted
features, are split into training, validation, and test sets. Various
machine learning algorithms, such as deep neural networks (DNNs),
convolutional neural networks (CNNs), and support vector
machines (SVMs), are evaluated to select the most suitable model
based on the dataset and problem characteristics.

5. Prediction: The trained models are applied to new voice Fig 2: Methodology flow diagram
recordings to predict the presence and type of respiratory diseases.

6. Detected Disease: The system can identify a variety of respiratory

diseases by analyzing voice recordings. Some of the common V. RESULT ANALYSIS
diseases include:
Advanced machine learning is used to derive the prediction results
• Asthma: Characterized by wheezing, shortness of breath, and of respiratory disease. techniques, specifically the Random Forest
tightness in the chest, which can affect voice quality. algorithm, which analyzes various clinical and audio features for the
identification of potential respiratory conditions. It processes audio
• Chronic Obstructive Pulmonary Disease (COPD): Includes signal such as cough sounds, breath sounds, or speech the algorithm
emphysema and chronic bronchitis, leading to persistent cough and extracts relevant patterns. That may indicate the presence of diseases
difficulty breathing. such as COPD, asthma, or pneumonia.
The system achieved promising results in the prediction of
respiratory diseases by using voice analysis:
• Accuracy: The accuracy for the Random Forest classifier
on the test dataset was 85%.
• Precision: The precision for COPD was 0.88, for asthma
0.82, and for pneumonia 0.90.
• Recall: The recall for COPD was 0.86, for asthma 0.80,
and for pneumonia 0.92.
• F1-Score: The F1-score for COPD was 0.87, for asthma
0.81, and for pneumonia 0.91.

Fig 3. Detection of Pneumonia

 survey. Inform. Med. Unlocked 2022, 29, 100832. [GoogleScholar]
[CrossRef]

[6]. Kim, H.; Jeon, J.; Han, Y.J.; Joo, Y.; Lee, J.; Lee, S.; Im, S.
Convolutional neural network classifies pathological voice change
in laryngeal cancer with high accuracy. J. Clin. Med. 2020, 9, 3415.
[Google Scholar] [CrossRef] [PubMed]

[7]. T., B.B.; Hee, H.I.; Teoh, O.H.; Lee, K.P.; Kapoor, S.;
Herremans, D.; Chen, J.M. Asthmatic versus healthy child
classification based on cough and vocalised /a:/ sounds. J.Acoust.
Soc. Am. 2020, 148, EL253–EL259. [Google Scholar] [CrossRef]
[PubMed]

[8]. Claxton, S.; Porter, P.; Brisbane, J.; Bear, N.; Wood, J.;
Peltonen, V.; Della, P.; Abeyratne, U. Identifying acute
exacerbations of chronic obstructive pulmonary disease using
patient-reported symptoms and cough feature analysis. NPJ Digit.
Fig 4: Detection of Asthma Med. 2021, 4, 107. [Google Scholar] [CrossRef]

Fig 5: Detection of COPD

VI. CONCLUSION

In summary, voice analysis-based and machine learning-based

prediction of respiratory conditions is a new achievement with 87-
91% toward non-invasive diagnosis. Due to meticulous
preprocessing of the voice data, feature extraction, and sophisticated
models with the training of those sophisticated models, it relies
significantly on the vocal characteristics' difference for detecting
conditions like asthma, COPD, bronchitis, and others. This will be
cost-effective and accessible alternative to traditional diagnostic
tools but also enables continuous monitoring and early detection,
which are crucial for timely medical intervention.

VII. REFERENCE

[1]. Md. Zahangir Alam, Albino Simonetti, Raffaele Brillantino,

Nick Tayler Chris Grainge, Pandula Siribaddana, S. A. Reza
Nouraei, James Batchelor, M. Sohel Rahman, Eliane V. Mancuzo,
John W. Holloway, Judith A. Holloway and Faisal I. Rezwan

[2]. Panagiotis Kapetanidis, Fotios Kalioras, Constantinos

Tsakonas, Pantelis Tzamalis, George Kontogiannis, Theodora
Karamanidou, Thanos G. Stavropoulos and Sotiris Nikoletseas

[3]. Alper Idrisoglu, Ana Luiza Dallora, Abbas Cheddad, Peter

Anderberg, AndreasJakobsson, Johan Sanmartin Berglund

[4]. Troncoso, Á.; Ortega, J.A.; Seepold, R.; Madrid, N.M. Non-
invasive devices forrespiratory sound monitoring. Procedia Comput.
Sci. 2021, 192, 3040–3048. [GoogleScholar] [CrossRef] [PubMed]

[5]. Ijaz, A.; Nabeel, M.; Masood, U.; Mahmood, T.; Hashmi, M.S.;
Posokhova, I.; Rizwan,A.; Imran, A. Towards using cough for
respiratory disease diagnosis by leveragingArtificial Intelligence: A