Time-Frequency analysis versus cepstrum analysis for validating respiratory sounds in infants and children

Automatic sound recognition for human body acoustic signals has attracted wide interests in recent years. However, the power of automatic sound recognition largely depends on the choice of features representing the acoustic signal. Recently, the time-frequency features and cepstral features are the most commonly utilized features in automatic recognition. The aim of this paper is to compare the time-frequency analysis versus cepstral analysis to find the best feature extraction technique. The one that has the greatest influence on the recognition and validation of diagnosed respiratory diseases in infants and children. This paper proves that the cepstral analysis of features result in better recognition accuracy, and the Mel-Frequency Cepstral Coefficients (MFCC) has the highest influence on recognition accuracy up to 94%, and more, versus the time-frequency features and linear cepstral technique. The used database was collected from infants and young children till the age of 12 years. This database comprised 492 disease of 3 different categories specifically rattle, stridor and wheeze.