The effects of ceiling height and absorber placement on speech intelligibility in simulated restaurants
The intelligibility of speech was measured in simulated rooms with parametrically manipulated acoustic features. In experiments 1 and 2 binaural room impulses were generated using a simple ray-tracing model for rectangular spaces. In order to simulate more complex geometries, including representations of furniture and room occupants, experiments 3 and 4 used CATT AcousticTM. The rooms were designed to simulate restaurant environments with either three or nine occupied tables. In Experiment 1, rooms of equal total absorbance were compared, but with most absorption located either on walls or on the ceiling. Wall absorption produced shorter reverberation times and improved speech reception thresholds (SRTs). In experiment 2, rooms differed in ceiling height. Lower ceilings produced shorter reverberation times but poorer SRTs. Both total absorbance and reverberation time were thus poorly correlated with speech intelligibility.
A psychoacoustic model of spatial release from masking (Jelfs et al., 2011) produced very accurate predictions of SRTs, based on the binaural room impulse responses for each experiment. Experiment 3 also varied ceiling height, but in combination with the effect of ground-level acoustic clutter, formed by furniture and room occupants. As predicted by the model, both high ceilings and acoustic clutter produced better SRTs. Experiment 4 compared acoustic treatments of the ceiling in the presence of the acoustic clutter. As predicted by the model, continuous acoustic ceilings were more effective at improving SRTs than suspended panels, and suspended panels were more effective if they were acoustically absorbent on both sides. The results suggest that the most effective control of reverberation for the purpose of conversational intelligibility is provided by absorbers placed vertically and close to the room occupants.