Towards measuring the Speech Transmission Index in fluctuating noise: Impulse response measurements

Introduction: The concept of the Modulation Transfer Function (MTF) in the field of room acoustics was introduced by Houtgast & Steen-eken (1973). The MTF can be used to predict speech intelligibility in stationary noise and re-verberation and can be expressed in one single value: the Speech Transmission Index (STI). The drawback of STI measurements using the classi-cal, direct method that it is time-consuming and that it is not validated for fluctuating background noise (IEC60268-16, 2011). However, the MTF as a result of reverberation can also be calculated ―indirectly‖ using impulse response measure-ments (based on Schroeder, 1978). Furthermore, when calculating the STI per time segment and applying the same time averaging as used by Rhebergen et al., 2006 (regarding the Extended Speech Intelligibility Index), the ―Extended‖ Speech Transmission Index (ESTI) can be calcu-lated in fluctuating noise. A prerequisite of using the method described here is that the impulse response can be measured reliably in fluctuating noise. In the current study we investigated acous-tical conditions under which the impulse re-sponse can be measured with sufficient precision to calculate the ESTI. Methods: Impulse response measurements were done by playing, recording and deconvolution of an exponential sweep signal. Background noise of a single speaker was simulated by playing the ISTS (Holube et al. 2010) through a loudspeaker. Experiment 1 was conducted in a room with vari-able absorption, different levels of background noise, and a fixed sweep level. Experiment 2 was conducted in a room with fixed absorption and background noise level, but with different sweep levels. Besides the 2 experiments, simulations with 6 different types of fluctuating noise were done in order to extrapolate the experimental findings to other acoustical conditions. Results: The experiments and simulations showed that a minimum broadband SNR of -5 dB in stationary noise was necessary to reliably es-timate the ESTI in 90% or more of the cases (a deviation of 0.015 STI units was allowed, based on IEC60268-16, 2011). In fluctuating noise a SNR of +5 dB was needed to reach the 100% criterion. Discussion and conclusion: The ISTS was cho-sen as a measurement signal due to its singletalker characteristics. The study could be extend-ed by using other noise types. We found that, when measuring the impulse response, a broad-band SNR of +5 dB was necessary to accurately calculate the ESTI. This is sufficient to perform the measurements in regular surroundings like classrooms and office floors.