Annu Haapakangas and Valtteri Hongisto switched on a masking system in a department of fifteen workers and measured before and after. Pink noise at 44 dBA came from ceiling loudspeakers, shaped to sound like ventilation, against an initial background of 36 dBA. The radius of distraction, the distance within which speech is intelligible enough to interrupt, fell from 15 metres to 7. Distraction by speech and by other varying office sounds both fell, noise-related stress fell, and the ratings of the acoustic environment improved. No negative effect appeared. Fifteen workers is a pilot and the paper calls itself one.
Haapakangas put numbers on that radius nine years later, pooling 21 workplaces and 883 respondents with room acoustic measurements taken to the ISO 3382-3 standard, and analysing them as individual participant data. Every additional metre of distraction distance raised the odds of high noise disturbance by about 13 per cent, and 14 per cent for disturbance by background speech specifically. Over five metres that compounds to about 84 and 92.5 per cent. Every decibel of speech level measured at four metres from the speaker raised the odds by 20 and 27 per cent. Of the 875 participants in the models, 38 per cent fell into the highly disturbed group.
Manuj Yadav and Densil Cabrera built a medium-sized open-plan office inside a climate chamber and simulated speech from spatially separated talkers. In the first experiment, with 60 participants, going from no talker to one cost 24 per cent of digit recall performance, and a second talker a further 12 per cent. In the second, with 62 participants, performance under two talkers stayed flat between 45 and 57 dB(A), which is the range almost every open-plan office sits in. Volume between 45 and 57 dB(A) made no difference to performance. The first audible talker cost 24 per cent and the second a further 12.
Schlittmeier tested the other half of that. Background speech at 35 dB(A) was compared with silence and with highly intelligible speech at 55 dB(A), across three experiments with 20, 24 and 28 participants. Soft and poorly intelligible speech restored verbal short-term memory and mental arithmetic performance almost to the level of silence. Soft and highly intelligible speech did not. Verbal-logical reasoning showed no effect of background speech at all. The subjective ratings did not follow the performance: soft and poorly intelligible speech was rated the least disturbing condition, and still more disturbing than silence.
Lee and Haan modelled four partition layouts, face to face, parallel, crossed-rod and orthogonal, at heights of 1.2, 1.5 and 1.8 metres, then ran an acoustic field experiment at 1.5 metres. The parallel and crossed-rod arrangements gave the best speech privacy at the lowest height. As the partitions got taller the differences between layouts disappeared, so the arrangement matters most where the partitions are low enough for people to see over.
Only one study here is field evidence at scale: 883 respondents across 21 workplaces, pooled by Haapakangas. The 2008 masking experiment ran in a single department of fifteen. Yadav tested 60 and 62 people in a climate chamber. Schlittmeier ran three laboratory experiments with 20, 24 and 28 participants. Lee and Haan had a computer model and one field measurement.