The Features Extraction of Infants Cries by Using Discrete Wavelet Transform Techniques

This article proposes the study and experiment of infant sound features extraction by using discrete wavelet transform (DWT) techniques. The DWTs using in this research are Haar, Symlet2 and Coiflet1 mother wavelets. In this classification, the Dunstan baby language is the infant sound data. The extracted features from the infant sound by DWT are learned by using the extreme learning machine (ELM) neural network. The results of this learning are compared in term of learning accuracy. From the experimental result, it is found that the average result of the ELM with Haar wavelet features extraction at number node of 30 is better than results of ELM with other wavelets in term of learning accuracy. However, there are insignificant differences in learning accuracy when the number of nodes is increased from 20 to 30 nodes.