Do speech-in-noise scores in normal-hearing humans correlate with amplitude modulation depth detection abilities?

Introduction
Extensive research has been performed to explore human temporal auditory processing abilities and their links to speech perception. This study aims to investigate the influence of temporal processing abilities, in form of amplitude modulation depth (AMD) detection abilities, on speech-in-noise recognition scores.

Methods
Seven young, normal-hearing adults participated in two psychoacoustic (PA1 and PA2) tasks and in a speech-in-noise test. Both PA tasks consisted of amplitude modulated and unmodulated noise presented monaurally to the left ear at 65dB SPL. The modulation frequency was set to 8Hz and the AMD was varied.
The PA1 paradigm determined behavioural AMD detection thresholds with a three-alternative forced-choice two-down/one-up task. The PA2 paradigm probed the AM detection ability for specific AMDs (10%-100%) and monitored the percentage of correct responses. Stimuli were presented individually and the participant determined whether it was modulated or unmodulated.
AzBio sentences were tested at three SNR levels (10dB, 5dB and 0dB) with respect to ten talker-babble noise. The percentage of correctly identified words was recorded as the speech-in-noise score.

Results
The PA1 AMD detection threshold indicating 70.7% correct responses ranges between 8.9% and 15.1% (mean of 12.49%) revealing overall similar discrimination abilities for this group.
Results for the PA2 task show high-pass characteristics with the drop-off located below 25% AMD. A one-way ANOVA revealed statistically significant differences between AMD detection scores (F(5,54)=37.78, p<0.001). Post-hoc Tukey comparisons determined that the 10% and 12.5% AMD conditions were significantly different from all other conditions (25%-100%) and from each other.
A one-way ANOVA revealed significant differences for the different SNR level speech test scores (F(2,33)=31.20, p<0.001). Following Tukey comparisons revealed that the 0dB SNR level is significantly different to the other conditions. Furthermore, a significant correlation was found between speech test scores at 0dB and PA2 scores for the 12.5% AMD level (r2=0.9077, p=0.0047), but not for the 10% level (r2=0.1626, p=0.7275) or the PA1 AMD detection thresholds (r2=0.0323, p=0.9440).

Conclusion
Speech-in-noise recognition scores correlate well with the ability to discriminate modulated and unmodulated sounds at 12.5% modulation depth, which also marks the average AMD detection threshold for this group at 70.7% correctness level. However, this finding has to be corroborated with an increased sample size. If validated, this method could be applied to evaluate or forecast speech recognition ability in clinical populations such as cochlear implant users.