What we talk about when we talk about speech intelligibility
When discussing speech intelligibility benefits, it is common to refer to the signal-to-noise ratio (SNR) where a listener’s ability to repeat the signal correctly 50% of the time (SNR50). If performance has been measured robustly, there should be objective equivalence in difficulty across any signal and noise pairs presented at a listener’s SNR50. It is reasonable to assume that the listener’s perception of difficulty will also be equivalent across stimuli presented at their respective SNR50s. We found this assumption of subjective equivalence to be false.
Twenty adult (median age of 67 years) listeners (nine female) of varying hearing ability (median better-ear average 29 dB HL) participated. In different blocks of trials, listeners first were tasked with repeating back IEEE sentences in same-spectrum or two-talker babble noise at various SNRs. Individual SNR50s were then estimated from the psychometric functions. Thereafter, in a modification to our previous method to measure the SNR JND, listeners heard on a given trial two intervals: a sentence presented in babble and the same sentence presented in same-spectrum noise. One interval would be presented at its SNR50 and the other at its SNR50 plus an increment varying from 0-8 dB in 2 dB increments. Listeners were asked to choose which sentence was clearer. All stimulus combinations and orders were counter-balanced and repeated 12 times.
The result of note was when there was a 0 dB increment (i.e., when both stimuli were presented at their SNR50). It was initially expected that listeners would choose each stimulus 50% of the time. Listeners on average, however, chose sentences in babble to be clearer 64% vs. 36% for sentences in same-spectrum noise [t(19) = 7.81; p < 0.0001]. In the rest of the conditions, this preference or “clarity gap” persisted. There was no correlation between the clarity gap and individual SNR50s nor individual differences in SNR50s. This particular result indicates a difference between objective, perceptual benefits and subjective, perceived benefits. If equivalent performance is not perceived as being equivalent in clarity across stimuli, perhaps an altogether different measure, such as effort, could yield subjective equivalence.
Funding — Work supported by the Medical Research Council (grant number U135097131) and the Chief Scientist Office of the Scottish Government.