Empirical Prediction of Speech Levels and Reverberation in Classrooms

Abstract

This paper discusses the development of empirical models for predicting total A-weighted speech levels and 1-kHz early-decay times in classrooms in an arbitrary state of occupancy. These are the two main quantities that affect speech intelligibility in classrooms. Three models for predicting early-decay time were developed. One was based on determining the contributions of various surface features to the average classroom-surface absorption coefficients. The other models, and those for predicting speech levels, were developed using multi-variable linear-regression techniques, and data previously measured in university classrooms or predicted empirically. By way of evaluation, the models were shown to re-predict the average values of the measured quantities in the original data-set with high accuracy, but they tended to underestimate the variability in the data. Predictions are presented to illustrate the performance of the models in the case of small and large hypothetical classrooms with low and high surface absorption, when unoccupied and occupied. The results are consistent with those measured in real classrooms. In particular, the speech-level model predicts physically-realistic decreases with distance from a speaker to a listener. The experimental data has also been used to determine typical ‘effective’ absorption coefficients for three classroom features – carpeted floors, absorbent ceilings and upholstered seating on carpeted floors – data which indicates the real-world performance that can be expected of these features, which may be useful in other prediction models and, for example, which provides information on the choice of treatments to meet the requirements of standards.