The dependence of diffusion parameters in a room acoustics prediction model on auditorium sizes and shapes

Abstract

The dependence of the calculation parameters that control the diffusion calculation in a room acoustics prediction computer model on auditorium sizes and shapes was investigated through a series of physical scale model experiments. Seven auditorium type scale models were used in total. The models had typical auditorium aspect ratios of around 2.5:1.8:1 and important auditorium features such as stage, balcony and raked seating. Their sizes varied from 5050 to 29 500 m3 and shapes included rectangular, fan, reversed fan, and hexagonal. The computer prediction model used a combination of randomized ray directions and secondary diffuse area source radiation to calculate diffuse reflections in its prediction of a room’s acoustic characteristics. The calculation process is controlled by two diffusion parameters: the diffuse-reflection coefficient of surfaces and a transition order which control the on-set of the diffusion calculation process. It was found that a diffuse-reflection coefficient of 0.1 was appropriate for the plain walls of the scale models in the 500- and 1-kHz octave bands, and this value did not change significantly with the scale models’ shapes and sizes. In the lower frequency bands the diffuse-reflection coefficient changed from 0.1 in the smaller scale models to over 0.4 in the larger models. A slight increase was also observed in models with more complicated shapes. For the transition order a low value (0) was found to be appropriate for the low frequency bands but a higher value (1 to 3) was required in the 500- and 1-kHz bands and also in situation where the specular/geometrical influence of a room was important. The average and root-mean-square prediction errors of the computer model using a generic set of diffusion parameters for the scale models were found to be, respectively, within 3% and 10% for RT, and 1.1 and 2 dB for sound level. For the clarity index C80, the prediction errors in the smallest model were within 1.9 and 2.2 dB, which were significantly higher than the 0.8 and 1.8 dB found in the other models.