Speech Envelope Reconstruction From Stereo EEG in a Speech Production Task

Significant progress has been made in decoding and synthesizing speech directly from intracranial brain recordings for speech neuroprosthetics. Much of this work is based on cortical recordings using microelectrode arrays (MEAs) or electrocorticography (ECoG). Unlike these cortical counterparts, stereotactic electroencephalography (sEEG) is minimally invasive, provides access to deeper brain structures, and is widely used clinically - facilitating access for conducting neuroscientific research. While clinical sEEG has some limitations for speech decoding research, these signals are envisioned to supplement the decoding achieved by cortical recordings. Here, we show that the acoustic speech envelope can be successfully reconstructed from sEEG recordings, with the strongest contributions from auditory and motor areas. This work further supports the viability of sEEG for speech decoding applications.