Audio recognition using feed forward neural network optimized by principle component analysis

Abstract

In our proposed model we have used PCA as dimension reduction technique and neural network for pattern recognition. Our goal was to recognize audios of two vowels spoken by Parkinson’s disease Patient. The vocal of these patients becomes unclear in later stage of the disease, therefore understanding them becomes difficult and hence our model is targeted to help them communicate. PCA was run to get the finest number of features to train the classifier. The classifier takes 30 percent of the feature to train and the rest 70% for testing and validation. Our model has yield a very high accuracy compared to other models.