UC San Diego

Speech is the basis of human communication and yet, the neural foundation of speech production and perception are still far from understood. Multiple proof of concept studies have demonstrated the potential to provide a fully closed loop speech prosthesis for people suffering from anarthria and other speech disabilities. In this thesis, we explore the value of intracranial electrophysiology to capture the variability of neural signals during vowel production and perception for speech prosthesis design. Using stereotactic electroencephalography, we analyze neural recordings to investigate how vowels are encoded in different brain regions and how vowel production correlates to neural activity. To enable these analyses, we first design audio signal processing algorithms and develop metrics to detect speech onset and to extract the first two formant frequencies from an input audio signal. We confirm that these two extracted formant frequencies can be used to uniquely identify vowels in the formant space. Second, we investigate the behavior of neural activity related to speech production by comparing it to that of silence and conclude that there exist significant differences between neural activity during speech and silence. Furthermore, we show that different brain regions respond differently to speech production revealing that there exists a spatially specific modulation relative to vowel production in the brain. This spatially distinct modulation has dynamics that correspond to vowel production onset time. In the final section of this work, we investigate the neural correlation to specific phoneme production and start to implement a classifier to decode vowel identity based on neural activity.