The use of neural network analysis to predict the acoustic performance of large rooms Part II. Predictions of the acoustical attributes of concert halls utilizing measured data

Abstract

A method of predicting the G values (the strength factor in dB), C 80 values (the clarity factor in dB) and LF (the lateral energy fraction) in concert halls has been investigated. Constructional and acoustical data for 72 unoccupied concert halls, in various countries, were utilized for the neural network analyses. One source in two positions, in 42 of the halls, one source in three positions, in 13 of the halls, and a single source in 17 of the halls were chosen, together with a combination of receiver positions were available to train the neural networks. In all cases, the source used was omnidirectional. Results show that for the position-dependent strength factor, G, the accuracy of the neural network predictions is within the subjective difference limen, which is 1 dB. (i.e. Δ E/ E=0.26 where E is energy density). Results also showed that the accuracy of neural network predictions for the clarity factor, C 80, and the lateral energy fraction, LF, are within the subjective difference limens of ±0.5 dB and ±0.05, respectively. Eight concert halls were used to assess the neural network analysis method — by comparing neural network predictions with measured values and with predictions using Barron’s revised theory. In addition, three of the eight halls were used to compare neural network results with those obtained using by the hybrid ray tracing computer model ODEON 4.01. The results indicate that there is a good basis for using trained neural networks to predict the acoustical attributes G, C 80 and LF at the schematic stage.