Predicting Acoustic Properties In Enclosure Using Neural Network

Abstract

This study aims to predict the reverberation time in enclosed space using neural network. The predicting of reverberation time in an enclosed space is conducted by designing five different enclosed space model equipped with windows, walls, floor, and ceiling in which the design process is completed by using Google SketchUp. The Neural Network training dataset from 5 different models of enclosed space with reverberation time at 500Hz were computed from ODEON 12.10. The Neural Network was conducted for 500Hz as absorption coefficient, volume of each model, and number of windows used as the input variable. Mean square error and Regression values were obtained to examine the neural network efficiency. Overall, the neural network efficiency shows a good result with Mean Square Error below 0.005 and regression above 0.9.