Detection of Common Cold From Speech Signals Using Breathiness Features

This study examines the classification of cold speech and healthy speech using breathiness measures. The breathiness features are crucial parameters for assessing voice quality since it provides information regarding speech disturbances. These breathiness measures include jitter, shimmer, glottal-to-noise excitation ratio, harmonic energy, harmonic energy of residue, harmonic-to-signal ratio, and number of voiced frames. The deep neural network (DNN) classifier is employed to evaluate the performance of breathiness features. The results indicate that breathiness features efficiently distinguish between healthy and cold speech classes. The performance for differentiating between healthy and cold speech is further enhanced by combining breathiness features with Mel-frequency cepstral coefficients (MFCC) features. On the develop and test partitions of the URTIC database, the combination of MFCC and breathiness features yields UAR of 68.21% and 66.50%, respectively.