MFCC Feature Selection for Infant Cry Classification

The infant calls for attention by crying. The crying comes from many reasons (e.g., attention, hunger, need to change wet diapers). The infant crying sounds seem too similar and difficult to classify by humans. This paper applies Mel Frequency Cepstral Coefficients (MFCC) for the feature extraction process. Various number of MFCC-feature are investigated and compared to obtain a suitable factor for classification techniques. Simple and well-known classification techniques (decision tree, naive Bayes, and support vector machine) are selected to classify the infant cry sounds. The support vector machine has highest performance metrics in both term of accuracy and F1score that are 70% and 71% respectively. Those are obtained from 11 features of MFCC. From the experimental results, MFCC:11 is suitable for infant crying sounds.