Automatic Cough Detection from Audio Signals Using Deep Learning and Hybrid CNN–SVM Models

¹ Faculty of Sciences Dhar Mahraz, Laboratory of Computer Science, Signals, Automation and Cognitivism, USMBA, Fez, Morocco
² SDIA Team, Multidisciplinary Faculty of Nador, Mohammed First University, Oujda, Morocco

^* Corresponding author: This email address is being protected from spambots. You need JavaScript enabled to view it.

Published online: 24 June 2026

Abstract

Coughing is a natural physiological reflex that helps maintain respiratory health by clearing the airways of irritants, fluids, and pathogens. It also serves as a key clinical indicator for various respiratory conditions, including asthma, infections, pulmonary diseases, and viral illnesses such as COVID-19[1–3]. This study focuses on the automatic detection of cough events from audio recordings using deep learning techniques. We developed and evaluated both convolutional neural network (CNN) and hybrid CNN-SVM architectures trained on multiple spectro-temporal feature representations, including Mel Spectrograms, Mel-Frequency Cepstral Coefficients, and Gammatone Frequency Cepstral Coefficients. The models were tested on a controlled and balanced dataset of cough and speech recordings, demonstrating near-perfect accuracy and high recall for cough detection. Comparative analysis highlights the effectiveness of CNNs for feature extraction and the added robustness provided by the CNN-SVM hybrid approach. These results establish a reliable methodological baseline for real-time and embedded cough detection systems, with potential applications in continuous respiratory monitoring and telehealth.