Alvarsson, Wiens, and Nilsson (2010) had 40 university students recover from four rounds of a stressful arithmetic task, each time with a different sound: a fountain mixed with birdsong, loud or quiet road traffic, or a quiet backyard. Skin conductance, a marker of sympathetic arousal, tended to recover faster during the nature sound than during the others, and it had the shortest estimated recovery half-life, though the average difference was significant only against loud traffic. Heart rate variability, their measure of parasympathetic activity, showed no effect.

Galbrun and Ali (2013) recorded a range of water features and played them over motorway traffic noise through headphones, asking listeners which sound was more peaceful and relaxing. Most of the water sounds lacked the low frequencies that dominate traffic noise, so they could not simply drown it out. Listeners still preferred stream sounds to fountains, and fountains to waterfalls, with the water at about the traffic level or no more than 3 dB below it. The study did not measure attention.

The "musical" quality of stream sounds comes from Minnaert resonance: air bubbles trapped in turbulent flow oscillate at specific frequencies based on their size, creating a naturally varied, non-repeating texture. We think that variety may keep the brain from tuning the sound out as quickly as it would steady noise; none of the studies we cite tested this.