20.09.2022 - Silvia Marchesotti

Brain-computer interfaces (BCIs) use neurophysiological signals to control an external device in real-time, without the involvement of the musculoskeletal system. During the past few years, technological advances in the field of BCI suggest that it is possible to decode speech related features directly from neural signals. These speech-neuroprostheses, once functioning, will provide a new channel of communication and rehabilitative solutions for patients that have lost the ability to communicate due to neurological disorders such as aphasia and locked-in syndrome.  

In this talk, I will review current speech-BCI systems, together with providing a background on the neural basis of speech production and current technologies used to acquire brain signals. I will present our current research activities aiming at tackling different aspects of real-time speech-decoding from non-invasive (electroencephalography, EEG) and invasive (intra-cranial EEG) recordings. In our experiments, participants imagine pronouncing syllables, chosen to elicit discriminable patterns of neural activity, and, based on the real time decoding of these signals, a visual feedback is provided. By performing this experiment with patients implanted with intracranial electrodes, we are identifying with high temporal and spatial accuracy the neural mechanisms involved in controlling the BCI. In another study, we have trained healthy participants to control the BCI over several consecutive days, with the goal of promoting learning and exploring whether is it possible for individuals who initially do not perform well to improve over time.  

In parallel, we are working on improving our BCI system from a pure technical point of view, developing adaptive rather than static classifiers typically used. By integrating in real-time the properties of the incoming EEG stream, the classifier is expected to compensate for the non-stationarity of EEG signals, potentially increasing decoding accuracy.  

Last, I will present current challenges in achieving effective speech decoding, trying to answer the question of whether speaking with the brain is nowadays -or will be in the near future- fact or fiction.