Calculating the speech transmission index in fluctuating noise: A data-driven approach in the short-term implementation

Everyday communication takes place in the concurrent presence of reverberation and background noise; the latter may have a fluctuating character and a speech-like spectrum, being for instance the result of multiple speakers talking together in the background (i.e., babble noise). The objective characterization of these listening conditions can be achieved by using a time-frame implementation of the Speech Transmission Index (STI) in the indirect scheme, named eSTI. One prerequisite of using the method is that the optimal time frame has to be determined. In this study, an experimental approach was used to determine the optimal time frame, defined as the one that provides coincident psychometric curves under stationary and fluctuating background noises. Matrixed-word listening tests were presented to 79 young adults with normal hearing. The speech reception task was presented under 26 listening conditions, created by varying signal-to-noise ratio, reverberation and noise type. By comparing the psychometric curves for the two noises, an interval of suitable frame durations was identified, ranging between 200 and 345 ms. Using a time frame within this interval thus ensures that the same eSTI value corresponds to the same predicted intelligibility, irrespective of the noise type.