Neural Signal-Based Speech Recognition: Leveraging EEG for Speech Detection

An electroencephalography (EEG) measures electri-cal activity in the brain via detecting signals being emitted by neurons using electrodes attached to the scalp. This technique gives an optimal amount of temporal but very little spatial coverage, i.e., it can record evoked or induced brain activity directly. This study intends to separately identify between-language differences in speech clarity, which may range on a spectrum from “crisp” (with occasional jarring moments but successful comprehension) to “confused”/unclear. It aims to differ speech clarity robustness and analyze EEG signals, respectively. We used data from earlier studies to reconstruct brain signals corresponding to speech clarity measured in EEG. In this work, we perform multi-level classification to enhance the EEG data analysis accuracy. This model achieved an accuracy of 90.48% during training, and can thus classify the speech clarity levels clear, unclear, more clear and very clear. This may provide a promising method to advance the evaluation of speech clarity with EEG. EEG, Electroencephalography, Speech.