An examination of ordering effects in a test to evaluate transient environmental sounds as heard indoors

Response data from a previously conducted test to evaluate transient environmental sounds as heard indoors were re-examined. The main goal was to determine to what extent subjects' evaluations of a sound were influenced by evaluations of sounds they had just heard. Another goal of the re-analysis was to examine annoyance models estimated from the responses of individual subjects and to compare them to annoyance models estimated from all subjects' responses. The test sounds were taken from recordings made simultaneously at two locations within a room. In Part A eighty sounds from one of the recording locations were played and subjects evaluated them, and in Part B they evaluated eighty recordings from the other location. The order of the playback was randomized with a different random order for each subject. Annoyance models of two types were estimated: Type I models are just based on sound quality metrics and Type II models are based on the annoyance rating of the previous sound and sound quality metrics. All possible combinations of three metrics from the group of considered metrics were examined for both Type I and Type II models and the best metric combination for each subject was determined. Also, metrics in the best model estimated from all subjects' data were chosen, and models with this fixed metric set re-estimated for individual subjects. Results from Part A and Part B are similar. Including the previous annoyance rating usually improved the coefficient of determination. In general, subjects whose best Type I models had very low coefficients of determination appear to have been more strongly influenced by the previous sound rating than other subjects whose best Type I models had a much higher coefficient of determination. For the latter group, there was also more consistency in which metrics appeared in their best models.