In 2010 Jesper Alvarsson gave 40 people a stressful mental arithmetic task and then played them either sounds from nature or sounds from noisy environments, with skin conductance recording throughout. Skin conductance fell to half its stressed value after 101.3 seconds under nature sound, 111.4 under low noise, 121.3 under ambient noise and 159.8 under high noise. Alvarsson is careful about what that shows. The difference reached significance only between nature sound and high noise, and the half-life values themselves could not be tested reliably.
Helena Jahncke ran the practical version of the question in 2016. Thirty students worked through a serial recall task in quiet, with one background voice, and with that voice masked three ways: by nature sound through headphones, by other voices through headphones and by headphones worn without any masking at all. A single voice was enough to damage performance. Nature sound was the only masker that protected it. Headphones alone did nothing. Perceived workload stayed high in every condition that had sound in it, so the method that worked was still not experienced as comfortable.
Arianna Latini built an office in virtual reality in 2024 and put 198 people through it, crossing three visual scenarios with three sound scenarios. At least 81 per cent called the traffic scenario chaotic and unpleasant. With natural sound the office was rated pleasant by 86 per cent, against 42 per cent for the office sound on its own. A second report of the same experiment puts the sound factor ahead of the visual one on cognitive performance, at 23 to 71 per cent against 12 to 39. That second report is a preprint and has not been through peer review.
Chao Liu looked for the neural counterpart in 2026 with 38 students across five sound conditions. Birdsong raised sound comfort, pleasure and satisfaction by 1.68 to 2.42 points against traffic noise, and water sound by 1.95 to 2.37. Alpha and beta power rose by 3.42 to 11.05 and 9.61 to 13.22 per cent, mostly over the frontal and parietal regions.
All four ran in a laboratory booth or in virtual reality. None of them ran in an office where people were doing their own work.