Research on Prediction Method of Objective Assessment of Building Acoustics Based on Machine Learning

Abstract The purpose of this study is to predict the objective assessment building acoustics more accurately and efficiently. In paper, neural network technology based on machine learning computer acoustic simulation are combined extract 10 typical characteristic parameters 3 target 800 halls rooms. Three matrix training sample databases established by using Odeon platform. reverberation time speech transmission index trained BP data fitting. R results in all than 0.95. MSE parameter range 0.01-0.05 STI less 1 × −4 . show that has good prediction accuracy, generalization application applicability. This method can quickly evaluate parameters, reduce manpower material resources, improve work efficiency.