The effect of stimulus choice on an EEG-based objective measure of speech intelligibility

Recently an objective measure of speech intelligibility, based on brain responses, has been developed using structured Matrix sentences as a stimulus. We investigated whether this method also works with more natural running speech as a stimulus, as this would be beneficial for clinical application and required for neuro-steered auditory prostheses. We hypothesized that because the syntactic structure of natural speech is less controlled and more linguistic top-down processing is involved, the outcome measure could be different using a natural story compared to the Matrix sentences.

We recorded the electroencephalogram (EEG) in 19 normal-hearing participants while they listened to two types of stimuli: Matrix sentences, 5-word sentences containing a proper name, verb, numeral, adjective and object with 10 options per word category presented randomly, and a natural story. Each stimulus was presented at different levels of speech understanding by adding speech weighted noise. To investigate the brain responses we analyzed neural tracking of the speech envelope because the speech envelope is known to be essential for speech understanding and can be reconstructed from the EEG in response to running speech. The speech envelope was reconstructed from the EEG in both the delta and the theta band with the use of a linear decoder and then correlated with the real speech envelope. We also conducted a test-retest analysis to asses the reliability of our objective measure.

For both stimulus types and filter bands the correlation between the speech envelope and the reconstructed envelope increased with increasing speech understanding. In addition, correlations were higher for the story compared to the Matrix sentences. These results indicate that neural envelope tracking is affected by the stimulus choice and it can be enhanced by the use of more natural speech and speech understanding. These findings suggest that the choice of the stimulus has to be considered based on the intended purpose of the measurement.

Acknowledgements
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 to Tom Francart). Further support came from KU Leuven Special Research Fund under grant OT/14/119. Research of Jonas Vanthornhout (1S10416N) and Eline Verschueren (1S86118N) is funded by a PhD grant of the Research Foundation Flanders (FWO).