Synthesizing EEG Signals from Speech Signals Using BCI's with EKF-Based Training

Abstract

A nonlinear oscillator differential equation model for EEG signals is linearized and discretized. The unknown parameters of this model are estimated using the EKF from noisy EEG signals. Likewise, the parameters of a linear difference equation model for speech are estimated using the EKF from noisy speech measurements. Based on the hypothesis that EEG and speech signals are correlated, a linear regression model that relates speech parameters to EEG parameters is proposed. By substituting this linear regression model into the speech differential equation, the EEG parameters for a fresh person are estimated from the speech signal alone using the EKF. This set up compares the generation of EEG data from speech data using a computer with training based on available EEG and speech signals from the BCI.