An experimental study of a time-frame implementation of the Speech Transmission Index in fluctuating speech-like noise conditions

Everyday communication takes place in the combined presence of reverberation and background noise, the latter having in some cases fluctuating characteristics and speech-like spectrum. To predict the speech intelligibility for fluctuating maskers, a time-frame implementation of the Speech Transmission Index (STI) in the indirect measurement scheme, named Extended STI (eSTI), has been recently proposed. Stationary speech spectrum noise is used as the probe signal and the a priori knowledge of the impulse response characterizing the transmission chain is required. A key issue of eSTI is the selection of the appropriate time frame for the calculations, whose duration is here assessed by a systematic experimental approach. Listening tests are developed for the scope, using speech-like stationary and fluctuating noises as maskers and rendering sound fields with various reverberation times and over a wide range of signal-to-noise ratios. An interval of frame durations providing equivalent values of eSTI under the two noises is identified. Within the target interval, statistically coincident psychometric curves are obtained for the two noises, thus ensuring that the same eSTI corresponds to the same speech intelligibility.