Detection of Common Cold from Speech Signals using Deep Neural Network.

Area of Science:

• Computational linguistics
• Speech processing
• Machine learning for healthcare

Background:

• The common cold, a viral respiratory illness, affects speech production by altering the vocal tract.
• Existing methods for cold detection are often invasive or lack real-time capabilities.

Purpose of the Study:

• To develop and evaluate a deep learning model for classifying speech affected by the common cold.
• To assess the model's performance against traditional machine learning classifiers.

Main Methods:

• Speech signals were analyzed using Mel-frequency cepstral coefficients (MFCC) and linear predictive coding (LPC).
• A deep learning model was trained on these features, with data imbalance addressed using SMOTE-Tomek links.
• Model performance was compared against logistic regression, random forest, and gradient boosted tree classifiers.

Main Results:

• The deep learning model achieved a higher Unweighted Average Recall (UAR) than the benchmark OpenSMILE SVM.
• The proposed method demonstrated comparable results to state-of-the-art techniques with reduced complexity and feature set size.

Conclusions:

• A novel, noninvasive deep learning approach for detecting the common cold via speech analysis has been successfully developed.
• This method holds potential for extension to the detection of other speech-affecting pathologies.