Detection of environmental sounds in informative backgrounds.

Abstract

In Gygi and Shafiro (2010), it was shown that the identifiability of environmental sounds presented in naturalistic scenes can be affected by the semantic content of the background scene: under certain conditions there is an incongruency advantage for sounds which are contextually incongruent, i.e., not semantically consistent with the background scene. Detection thresholds for the same set of sounds and scenes were obtained using adaptive tracking. The 28 background scenes were common auditory settings, such as a street scene or a restaurant. The 194 individual target sound tokens represented 50 distinct sound sources (e.g., baby crying, whistle blowing, etc.) and were presented in scenes that were both congruent and incongruent. The results show a wide range of detection thresholds, from ∼−10 to ∼−35 dB sound to scene ratio. These were overall much lower than detection thresholds for a similar corpus of sounds in broadband noise in Gygi et al. (1998), showing that naturalistic backgrounds can allow for greater detectibility; however, there was no significant difference between the thresholds in the congruent and incongruent settings, indicating that, unlike for sound identification, the semantic fit of the sound and scene was not a major feature in the detectibility.