Background: In previous work, we found that the pure-tone average (PTA) had only a minimal effect on speech perception of the digits-in-noise (DIN) test when speech was presented well above pure-tone thresholds. However, a substantial association between PTA and DIN was found when the speech intelligibility index (SII) was used to account for audibility rather than level- and spectral adjustments during stimulation. This relationship may be driven by residual audibility effects that are not fully accounted for by the SII.

Rationale: Accurate audibility estimation is essential for investigating factors that influence speech understanding, such as cognition and suprathreshold processing. Adjusting for audibility ensures that differences in speech understanding can be attributed to factors of interest rather than to variations in hearing sensitivity.

Methods: Speech reception thresholds (SRTs) were measured with three DIN test variations. From 2011 to 2014, 3455 participants had a measurement with a fixed noise level of 55 dB SPL (DIN55), and from 2014 to 2016, 1513 participants were measured at 70 dB (DIN70). Of these, 711 also completed the DINvar test, which ensured audibility with spectrally adapted speech. We used the SII in two ways. SIISRT calculated the audibility of the speech-in-noise signal, while SIIspeech reflected audibility of the speech relative to hearing thresholds. Given the non-linear relationships between the SRT/SII and PTA, Generalized Additive Models (GAMs) were used to allow flexible modeling by fitting smooth functions.

Results: The relationship between SIISRT and PTA was stronger for DIN55 (Radj2 = .16, F(5.09, 6.13) = 103.8, p < .001) compared to DIN70 (Radj2 = .11, F(4.41, 5.44) = 32.8, p < .001). With DINvar, a weak relationship between SRT and PTA was found (Radj2 = .01, F(2.57, 3.25) = 3.55, p = 0.01). Among participants with valid DINvar measurements, of which PTAs are limited to 40 dB HL, the SIISRT and SRT of the DIN70 test were nearly perfectly correlated (Radj2 = .97, F(4.1, 5.12) = 4342, p < .001). We validated this with DIN55 for participants with a similar maximum PTA level relative to the sound level (Radj2 = .94, F(5.42, 6.6) = 5900, p < .001). After a linear correction with the SIISRT with the SIIspeech in these subsamples, only weak relationships with PTA were found (DIN55: Radj2 = .07, F(2.43, 3.07) = 62.91, p < .001 and DIN70: Radj2 = .02, F(1, 13.64) = 13.64, p < .001).

Conclusions: The SII may underestimate audibility effects in speech-in-noise perception. The relationship between SIISRT and PTA could be explained by the decrease in speech audibility relative to hearing thresholds at higher PTAs. The SII model should be adapted or extended to incorporate this audibility effect.