Classroom speech intelligibility prediction using backpropagation Neural Network

Abstract

In terms of individual communication, speech is the most important and efficient means, even in today's multi-media society. Thus, classrooms are mainly used for delivering speech between lecturers and students, it is important that acoustic designs accommodate and enhance such use. In achieving the highest possible speech intelligibility, the acoustical design of classrooms should be based on all the listeners in the classrooms. This paper investigates the effect of Signal to Noise Ratio (S/N) on Speech Transmission Index (STI) in University classrooms. The sound pressure levels are measured at different classrooms positions. Based on the measured speech levels, STI at various listeners' positions are determined and a simple backpropagation network model is developed to predict the STI at various listeners' positions and for various speech levels.