Factors affecting the subjective impression of speech intelligibility

From laboratory investigations to clinical interventions, we often assume that our objective measures of speech intelligibility have perceptual relevance. By the same assumption, an objective equivalence, such as the signal-to-noise ratio (SNR) where 50% of the keywords in a sentence can be correctly repeated (SNR50), should have perceptual equivalence, such as being judged to be equally clear. We previously found, however, that when presented with a sentence in same-spectrum noise (SSN) and two-talker babble at their respective individual SNR50s, participants were approx. 30% more likely to choose the sentence in two-talker babble being clearer, despite being at a lower, objectively equivalent SNR50. In the current study, we explore whether this clarity bias persists under different conditions, and if it is due to (1) the presence of speech in the masker, and/or (2) unmasked segments of the target signal that do not contribute to speech intelligibility. Thirty-six adults of varying hearing ability are first asked to repeat back sentences presented at various SNRs in three noise types: SSN, n-talker babble (n = 1, 2, 4, 8 & 16), and SSN modulated with the envelope (low-pass Hilbert transform) of n-talker babble. Individual SNR50s are estimated for each of the 11 noise conditions. Participants are then asked which of a pair of stimuli presented at their respective SNR50s are clearer. Pairs consist of either (a) n-talker babble and SSN, (b) n-talker modulated SSN and SSN, or (c) n-talker babble and n-talker modulated SSN. If clarity bias is due to speech in the masker, the bias should be reduced when the carrier signal is noise modulated by the babble. If clarity bias is due to unmasked segments, the bias should persist with a babble-modulated noise carrier, but decrease with increasing number of talkers in the babble. Results will be analysed for both objective and subjective disparities, and discussed in terms of the effect of noise type on subjective reports of speech intelligibility.

Funding: This work was supported by the Medical Research Council [grant number MR/S003576/1]; and the Chief Scientist Office of the Scottish Government.