Language is ingrained in our thought process and consciousness. It takes several forms, of which
speech is the most spontaneous. Oral communication is essential to exchange information,
emotions, and abstract ideas. Being unable to speak cuts one away from society, causing great
suffering for the patient and his relatives as well as a complicating clinical care. People that still have
an intact cerebral function but can’t communicate, like patients with neurological disease – cerebral
palsy, stroke or spinal cord injury and locked-in syndrome for example (2) – would greatly
beneficiate from speech decoding devices. However, brain computer interfaces (BCIs) aren’t at that
stage to this day. Improving our understanding of how our brains treat language as well as refining
decoding of speech-related activity will push BCIs into a new era, changing the quality of life of
patients suffering from these debilitating conditions for good. To this day, different brain-waves
patterns are known (3). Sorted by increasing frequency from delta to gamma, their role starts to be
discovered (4, 5). Theta waves – ranging from 4 to 8 Hz – have been shown to be essential for speech
intelligibility (6) and to correlate with speech signal envelope (7). In this project, the involvement of
theta waves in listening and speech production tasks was explored further. Theta waves signal as
well as signal envelope were explored in different speech-related conditions and their potential
correlation to speech onset and offset was explored.
Electrocorticogram (ECoG) recording of brain activity of four patients undergoing surgery for severe
epilepsy were obtained. ECoG is little invasive and offers an excellent time and space sensitivity.
Acquired data encompasses three conditions in which patients listened, imagined, or repeated words
or sentences. This data and the concomitant audio recording were analysed using Matlab®. ECoG
signal was filtered for theta range (4-8Hz), the envelope was taken, and both signal and envelope
were normalised to detect significant changes. Brain heat maps were then computed to assess
spatio-temporal changes in theta activity and average changes over the different tasks. Finally, to
assess the potential of theta activity as speech onset and offset biomarker – an important feature to
improve speech decoding BCIs –, the correlation between theta waves and audio recordings was
investigated. A linear regression model linking different inputs (theta signal, theta envelope) to
outputs (audio signal, audio envelope, trigger values – “ongoing task markers”) was tested.
Results indicate that theta activity and theta envelope is significantly modulated in listening-related
tasks. Marked decreases in theta activity and envelope were observed in all experimental set-ups,
with focal increases in theta activity in the left primary auditory cortex during listening. Average
theta envelope in tasks in which subjects were to listen or repeat words was modulated with regard
to baseline in both hemispheres. Theta envelope correlated well to audio envelope and “markers of
ongoing task” in these cases but less so when tasks involved imagined and overt speech.
Thus, theta envelope seems to be a good marker of listening tasks but appears to be a poor marker
of speech intention.