The speech intelligibility index assumes that the dynamic range of speech is 30 dB, but it is unclear what the empirical basis of this assumption is. Leclère et al. (2016) tested the range by using speech-in-noise stimuli for which the speech-to-noise ratio was strongly frequency dependent. The speech or the noise was filtered using a step-function filter that attenuated the level by a specified number of decibels above or below 1400 Hz. The results showed that as the size of this step change was increased, the speech reception threshold (SRT) changed progressively up to a certain level of attenuation and then stabilised. The range of attenuations over which changes were significant was around 40 dB, indicating that up to this point, the signal-to-noise ratios both above and below 1400 Hz were affecting intelligibility. This result suggests that 30 dB may be an underestimate of the dynamic range relevant to intelligibility. The present experiment took an analogous approach using temporal modulation of the masker. Interrupted noise reduces SRTs by around 20 dB compared to continuous noise of equal long-term intensity. Rather than fully interrupting the noise, it was subject to square wave modulation of controlled depth. Consistent with Leclère et al. (2016), increases in modulation depth up to 40 dB produced progressive changes in SRT. At 40 dB attenuation or greater, the full 20-dB effect of interrupting the noise was observed.

References: Leclère, T., Théry, D., Lavandier, M. and Culling, J. F. (2016) “Speech intelligibility for target and masker with different spectra” in van Dijk, P., Başkent, D., de Kleine, A., Gaudrain, E., Lanting, C. & Woldhuis, R. (eds) Physiology, Psychoacoustics and Cognition in Normal and Impaired Hearing (Springer), 257-266.