Predicting Speech in Noise Perception using Measures of Neural Entrainment
Speech perception ability in a cocktail party environment is highly variable, even for people with clinically healthy hearing. The mechanisms of these differences are poorly understood. A possible reason for these differences could be due to the neural mechanisms of auditory attention. This investigation aims to quantify how well a person is able to neurally entrain to a continuous narrative using a temporal response function (TRF), then evaluate if that TRF measure is correlated with an individual's performance on a selective attention and narrative comprehension task. This was accomplished using a cohort of twenty-five young and healthy listeners who completed speech-in-noise (SIN) perception tasks while their neural activity was recorded with EEG. This EEG data and regressors (speech envelope and gammatone spectrogram) of an auditory stimulus was used to train TRF models. TRF cross validation correlation coefficients were then used in multiple regression models to predict a participant's SIN performance. Task performance had a significant positive relationship with the neural entrainment to an attended stimulus and a negative relationship with the neural entrainment to an unattended stimulus. This was observed with both the selective attention and narrative comprehension measures. The difference of the attended and unattended correlation coefficients were also significantly related to increased task performance. These measures could be a potential biomarker to predict an individual's speech perception in a noisy environment.