Acoustical designing for speech quality in theaters

Speech quality, a subjective measure of overall excellence determined by preference testing, was originally developed for evaluating speech system designs. The application of speech quality measurements to the evaluation of auditorium acoustics design required two models: a real-time multichannel electroacoustic model to simulate physical sound fields, and a linear/scalar products mathematical model to analyze subjective preferences (together with an analysis of variance method for determining statistical significance and scalability levels). In the present experiments, speech quality is shown to be either multidimensional or unidimensional in configuration, and valid at either interval or ordinal levels of measurement. Four consensus dimensions (reflections ratio; excessive reverberation; discrete echo disturbance; naturalness) and three individual differences dimensions (reverberance; spatial coloration; evenness of echogram) were identified. The objective index, signal-to-noise ratio, gave a high correlation with speech quality for simulated sound fields without excessive background noise. [Work supported by the SERC, United Kingdom.]