Speaker
Description
Comprehensive assessments of room acoustic quality rely on perceptual evaluations. Conducting such evaluations online avoids time-consuming laboratory experiments at the expense of experimental control. This seems hardly feasible, given the technological demands of the accurate reproduction of room auralizations. However, since playback system deficiencies differ between subjects, their individual errors would just add noise to the inevitable measurement error, which might be counterbalanced by including more participants. This study therefore compares perceptual ratings of six rooms using the Room Acoustical Quality Inventory (RAQI), obtained from two laboratory studies with 40 and 31 participants, respectively, and an online study with 121 participants. The two laboratory studies employed either dynamic binaural renderings, or a static auralization with fixed head orientation, which was also used in the online study. Bayesian hierarchical regression models based on Multivariate Generalizability Theory were used to quantify measurement error attributable to the participants and to compare true score estimates of the measured RAQI factors. Preliminary results indicate that differences in the estimated true score, participant error variance, and score reliability depend on the specific RAQI factor and are largest within the Quality and Reverberance factors. The results of the laboratory experiment with fixed head orientation showed differences to the dynamic condition that were comparable to those of the online study. While a more diverse set of evaluated rooms would be needed for more conclusive results, the proposed method allows for a sophisticated assessment of measurement error in online experiments, instead of ruling them out from the outset.