Neural network models for speech inteligibility assessment in university classroom

Abstract

Adequate speech intelligibility should be the primary goal in acoustical design of classrooms. Typical design parameters are reverberation time and background noise level. However for predicting the Speech Transmission Index (STI) of a room, the designer should have knowledge about the early decay time at various listener's position in a classroom. In this paper, simple neural network models are developed to predict the STI and speech clarity (SC) of a classroom at various listeners' position based on the speaker's sound pressure level. Based on the network model, the variation of STI and SC at various listeners 'positions are mapped.