Estimation of BMI status via speech signals using short-term cepstral features

Fatness is a serious health problem worldwide because of the danger factors associated with diseases which cause permanent psychological effect. To classify normal weight, overweight, underweight and obesity, body mass index (BMI) is the most recognized and extensively used measurement. BMI measurement has its limits in some cases like overstatements in athletes, and underestimates in elderly. Thus, the paper reports the estimation of BMI (body mass index) status via speech signals using the short-term cesptral speech feature extraction methods, Mel-frequency cepstral coefficients (MFCCs) and its deltas (Delta and Delta-Delta) and Linear Prediction Coding (LPC) based Cepstral parameters (LPCs, linear prediction cepstral coefficients -LPCCs and weighted LPCCs). Two different classifiers, probabilistic neural network (PNN) and k-nearest neighbor (KNN) were used for the classification of the three BMI statuses (normal, overweight and obese). The 10-fold cross validation method was used to validate the reliability of the classifier results. PNN gives the best average BMI status classification accuracy of 87% using the short-term cepstral features.