Auditory detection of water sounds in a natural habitat: Psychophysical and modeling data

This study explored the ability of humans to detect water sounds within natural scenes. Here, we assessed the detection performance of a stream sound in a Mediterranean forest. Acoustic samples were recorded at six distances (0–200 m) from a river, at three moments of the day and two seasons. 2-s long samples were presented diotically to 3333 participants tested online using a 2-interval, 2-alternative forced-choice detection paradigm. On a typical trial, listeners were presented in random order with a target interval playing a sample recorded between 0 and 100 m from the river and a comparison interval playing a sample recorded 200 m away from the river and then asked to report the interval containing water sounds. Each participant completed a single trial only. Water detection performance depended on distance and season only. Water detection was then simulated using an auditory texture model using the same task and stimuli. The model replicated the main trends found in human data when basing its decisions upon skewness and kurtosis of amplitude envelopes and the correlation of envelope fluctuations across modulation bands. Together, these results suggest that humans detect water in natural environments on the basis of sparseness and coordination of temporal-envelope cues. [Work supported by ANR AUDIECO/ALEAU.]