Cortical Signatures of Heard and Imagined Speech Envelopes.

摘要

We performed an experiment with both heard and imagined speech to determine whether cortical signatures of heard speech can be used to identify imagined speech. Each trial in the experiment presented one of six possible spoken sentences; it was both heard and, immediately afterwards, produced in imagination. The analysis focused on the use of envelope following responses (EFRs) to identify sentences. Source imaging methods were used to find the cortical origins of EFRs to heard speech. Reconstructing the EEG from the strongest sources of the EFRs in parietal and temporal cortex improved the correlation between EEG and the amplitude envelope of the heard speech. Single- trial classification performance was statistically significant for two of eight subjects. Significant classification performance was found for all subjects when one used EEG data from multiple trials of the same sentence, concatenated to produce data of greater duration. Activities at the cortical sources determined for heard speech were estimated from EEG data recorded while speech was imagined, in order to classify the imagined speech. Classification performance improves as the duration of EEG data increases; about seven trials of the same sentence are required for classification of the imagined sentence to reach statistical significance. These results suggest imagining speech engages some of the cortical populations involved in perceiving speech, as suggested by models of speech perception and production.