Performance comparison of the artificial neural network and the k-nearest neighbor classifiers in classroom speech intelligibility prediction application

Abstract

Classroom speech intelligibility has become one of the major concerns in education nowadays. In any classrooms and educational facilities, an optimal speech intelligibility level is required to ensure that the listeners at any location in the classroom have a good perception of the message that is conveyed by the speaker. Classrooms need to be designed carefully in order to give optimal classroom speech intelligibility level. In this paper, two different types of artificial intelligence methods are proposed to implement the prediction application: Artificial Neural Network (ANN) and the k-Nearest Neighbor (k-NN). Both classifiers are trained separately using the previously acquired datasets which consist of acoustical parameters and the speech intelligibility of actual classrooms. Results show that the ANN performs better on imbalanced datasets compared to the k-NN.