MENOnet JOURNAL: WORKS in PROGRESS in EMBEDDED COMPUTING (WiPiEC), Volume 7, Issue 1, JUNE 2021 1 Utilization of Low-Cost Sound Sensors with a built in Microphone as a Respiratory Pattern Sound Indicator and a Risk Mitigation Tool In response to COVID-19 Gordana Laštovička-Medin University of Montenegro Podgorica, Montenegro gordana.medin@gmail.com Rajka Pejanović University of Montenegro Podgorica, Montenegro p.r.rajka@gmail.com Abstract—Since the detection of pattern abnormalities may lead to not only the prevention of chronic respiratory diseases but also other diseases, many techniques have been developed in order to detect breathing and coughing patterns. To benefit from the cross- disciplinary studies we have decided to expose physics students to both: learning about sound using coughing as a targeted research topic and to develop a demo tool that is useful for building on exploratory skills and provides them with solid knowledge for future more advanced scientific research in biomedical engineering. A low-cost microphone sensor was tested for the purpose of understanding whether it can be used not only as a sound indicator but more broadly as a risk mitigation tool during a pandemic such as the current pandemic, COVID-19. The final goal of this long-term project is to build mathematical models aiding the identification of features from sound samples and to apply a classifier algorithm based on the machine learning technique at the final stage of research. Keywords- COVID-19, sound pattern, human generated cough and breathing, low-cost sound sensor, loudness I. INTRODUCTION COVID-19 is a respiratory infection caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) [1]. Prominent symptoms of COVID-19 include coughing and breathing difficulties. Cough sound analysis helps us to differentiate two similar sounds and to define the objective correlations with spirometry and clinical diagnosis [2], including Cough Peak Flow using cough sounds [3]. The auscultation of the respiratory system is another diagnostic technique and an inexpensive, noninvasive, safe, and easy-to-perform method [4]. The parameters such as frequency, intensity, and timbre of sound are of particular interest for the classification of respiratory diseases and are defined as follows. Pitch is the subjective perception of sound's frequency and depends on the frequency while amplitude of loudness is related to the energy of sound waves and is measured by the height of sound waves from the mean position; it is the subjective perception of amplitude. Quality or timbre is an important property of sound that differentiates two sounds with the same pitch and loudness. The fundamental frequency or primary frequency is the lowest frequency of a sound wave and it determines the pitch of the sound; the frequencies higher than the fundamental frequencies are called overtones while harmonics are overtones whose frequencies are whole number multiples of the fundamental frequency. However, real-world sounds are not usually deterministic: they do not just have simple harmonics of the fundamental frequency. Instead, they also have unpredictable “inharmonic” frequencies that are not structured as noise. Thus having complete understanding of these measurable quantities and designing an experiment where those features are not lost when recorded and processed is crucial for further applications of machine learning techniques [5]. The issue is also how to deal with the research complexity without compromising the flexibility of techniques required for the extraction of sound features and still providing a comprehensive outcome that would not compress important information for the sake of data reduction. Towards this, this paper presents an early effort, mostly exploratory based, in building the capacity for such a complex and comprehensive task and towards creating a cough/respiratory sound database in Montenegro. II. EXPERIMENTAL PROCEDURE A. Research question The cough frequency is the most basic measure of coughing, but the objective study of cough signals has the potential to identify further features which may be clinically relevant and hence useful endpoints to study. Here we recall an early measurement (Fig. 1a) that used the audio tapes as research tools where the behavioral changes were monitored in order to extract untypical patterns over a longer time period [6]. Quantifying a cough was never an easy task, it is still not fully understood, and the symptoms are often incorrectly assigned. There is also no universally agreed unit of cough. The most intuitive way to