Dual-Tree Complex Wavelet Transform for the Automatic Detection of the Common Cold Based on Speech Signals

The acoustic and prosodic features of speech change in the presence of various health states. Biomedical engineering has enormous promise for developing non-invasive diagnostic technologies that use voice as a modality. The common cold is a highly prevalent sickness that affects a significant proportion of the global population throughout the year. The utilization of speech signals for the detection of the common cold has experienced a surge in popularity in recent times. In this study, the dual-tree complex wavelet transform (DTCWT) based new feature extraction technique is proposed for diagnosing common cold infection. First, we have employed the DTCWT to break down the speech signal into many sub-band coefficients. Then the features such as mean, variance, skewness, kurtosis, energy, approximate entropy, Renyi entropy, and permutation entropy are extracted from these sub-band coefficients. The URTIC database is utilized to assess the effectiveness of the proposed features. The classification results achieved using the transformer model show that the proposed algorithm detected cold from a speech sample with UAR of 68.66% and 64.52% on the develop and test set of the URTIC dataset. We have obtained comparable results with the state-of-the-art methods. The DTCWT captures subtle changes in speech signals, making it well-suited for detecting common cold symptoms. Its ability to provide both time-frequency localization and phase information enables it to discriminate between healthy and cold-affected speech patterns, leading to improved classification accuracy.