Predicting speech understanding from EEG recordings: effect of attention

Objectives Measuring speech understanding is currently done with behavioural tests. Behavioural tests require the active participation of the test subject and are strongly influenced by the knowledge of language, cognitive skills and attention. We aim to develop an objective and automatic correlate of speech understanding using EEG measurements. In this study we evaluate this electro-physiological measure to examine if speech signals in noise at different signal-to-noise ratios (SNRs) elicit systematically different responses and thus provide the building blocks to estimate the speech reception threshold objectively. We also investigate the effect of attention on the objective measure, to assess its suitability for clinical purposes. Methods 18 normal-hearing adults participated in a first experiment. The subjects listened attentively to the Flemish Matrix speech material combined with speech weighted noise at different SNRs (-12.5 dB until -1 dB) while their EEG was recorded. By applying a decoder that estimates the stimulus envelope based on the EEG data and correlating it with the original speech envelope, we obtained an estimate of neural entrainment of the speech envelope, which we hypothesise to be a prerequisite for speech understanding. By choosing the investigated frequency range and integration window of the decoder we were able to find monotonically increasing speech entrainment as a function of stimulus SNR. In a second experiment, 5 normal-hearing adults did the same task while their attention was being distracted by a film or video game. This scenario enabled us to investigate the effect of attention on envelope reconstruction. Results In the high SNR conditions the estimated envelope was better correlated with the envelope of the clean speech stimulus than in the low SNR conditions, meaning that the stimulus was better represented in the EEG data and thus presumably better understood. Optimal decoder parameters were delta band (0.5 Hz – 4 Hz) and an integration window of 0 to 75 ms. When the attention of the subjects was being distracted, the stimulus envelope was still better preserved in high SNR conditions. Conclusion Using EEG measurements we were able to estimate the SNR of a presented stimulus even when the attention of the subject was distracted. This is the first step towards automatically and objectively estimating speech perception measures, such as the speech reception threshold. Acknowledgements Financial support for this project is provided by the KU Leuven Special Research Fund under grant OT/14/119. Research funded by a PhD grant of the Research Foundation Flanders (FWO). This project has received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (grant agreement No 637424), and from the YouReCa Junior Mobility Programme under grant JUMO/15/034 to Jonas Vanthornhout.status: publishe