In-situ measurement method of speech privacy focusing on statistics of acoustic features of speech and background noise

Abstract

The evaluation of speech privacy in architectural spaces is done with a kind of transfer function of the space, such as the slope of the distance decay of the sound pressure level (SPL) and the level difference between the two rooms. From these functions, it is possible to estimate the speech-to-noise ratio at evaluation points, and then we can evaluate the degree of information leakage and work disturbance. However, the above indices are obtained from acoustic measurements or numerical simulations by experts, and can not be measured in the target space under operation. In this study, methods for estimating the speech-to-noise ratio from only the intruding speech recorded at the measurement point were investigated. We focused on steep changes in the SPL and the strength of the harmonic structure as acoustic features of speech that are considered to remain even at low speech-to-noise ratios. These features were quantified for each short time window, and the percentage that exceeded a threshold predetermined from statistics of a steady-state noise was obtained as an index for the estimation. The proposed index corresponded with the speech-to-noise ratio in both simulations and measurements.