Stimulus-brain activity alignment between speech and EEG signals in cochlear implant users, more than an artifact?

A number of studies (e.g. Luo & Poeppel, 2007) suggest that synchronization between neural oscillations, as they are observed between different bands of electroencephalographic (EEG) signals, and the temporal amplitude modulations of speech is foundational to speech processing (Giraud & Poeppel, 2012). Speech intelligibility appears to vary with the strength of such alignment between the brain and acoustic signals. These studies are based on vocoded speech materials, and hence stimuli that are similar to the signal transmitted via cochlear implants (CI). EEG studies with CI users suffer from the presence of electrical artifacts induced by the device. This study investigates the phase alignment between EEG signals recorded with CI users and envelopes of natural sentence stimuli, and queries how much such alignment reflects brain signals engaged in speech processing or a CI induced artifact.

EEG signals within the theta range (3-8 Hz) of eight CI users with their own device, and eight normal hearing (NH) participants, recorded while they were listening to naturally spoken sentences, were compared in terms of their alignment with the signal envelopes. The analysis involved a cross-correlation between the envelopes and the EEG channels to extract the lag between the signals. Coherence between aligned signals was then measured in terms of correlation and phase coherence. Preliminary results show for CI users correlations between 0.29 and 0.69, with higher values observed for channels close to the CI but also for contra-lateral temporal and central channels. For NH listeners, correlations were found on left temporal and central channels (0.16-0.33). The EEG signal lagged behind the speech signal for 120 ms to 380 ms for CI users, and for 200 till 450 ms for NH listeners. The correlation between speech envelopes and signals recorded in their respective trials was generally greater than the correlation found between EEG signals correlated with randomly chosen speech envelopes.

Greater coherence between the speech signal and the channels in vicinity to the CI, together with the absence of coherence for these channels in NH listeners, suggest that signal transmitted via the device is at the source of this alignment. The question of whether greater coherence reported for NH listeners in studies with vocoded stimuli reflects the less natural amplitude modulations in these signals is currently being tested. The coherence found for the central channels for NH and CI listeners, however, suggests that this alignment may indeed reflect a step in speech processing.