Investigation the Capability of Neural Network in Predicting Reverberation Time on Classroom

Musli Nizam Yahya, Toru Otsuru, Reiji Tomiku, Takeshi Okozono


The purpose of this paper is to investigate the capability of neural network in predicting a classroom’s reverberation time. A classroom in Oita University was chosen as a sample to obtain the virtual data (reverberation time) based on 20 types of sound absorptions coefficients using Finite Element Method (FEM) and Sabine equation. The capability of FEM has shown that it is able to simulate virtual data in each location of a classroom. To develop a neural network model, virtual data (721 data) was taken from FEM for the learning process. The assessment was made by using testing subset (20% from 721 data) to verify the performance. The testing’s means square error (MSE) was 3.7751×10-4 and correlation coefficient (R2) was 0.992 approximately to 1. The optimum network used was 4 hidden nodes. Extended assessment was made using the unseen data (35 data) and it showed that neural network prediction was approximately close to the actual data with MSE is 4.154×10-4. Basically, the capability of reverberation time prediction using neural network is shown in this paper.



Classroom; Reverberation time; Neural network; Finite element methods (FEM);

Full Text:


Copyright (c)

ISSN: 2180-3242

Creative Commons License
This OJS site and its metadata are licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.