Spatial and Spectral Fingerprint in the Brain: Speaker Identification from Single Trial MEG Signals

Brain activity signals are unique subject-specific biological features that can not be forged or stolen. Recognizing this inherent trait, brain waves are recently being acknowledged as a far more secure, sensitive, and confidential biometric approach for user identification. Yet, current electroencephalography (EEG) based biometric systems are still in infancy considering their requirement of a large number of sensors and lower recognition performance compared to present biometric modalities. In this study, we investigated the spatial and spectral fingerprints in the brain with magnetoencephalography (MEG) for speaker identification during rest (pre-stimuli) and speech production. Experimental results suggested that the frontal and the temporal regions of the brain and higher frequency (gamma and high gamma) neural oscillations are more dominating for speaker identification. Moreover, we also found that two optimally located MEG sensors are sufficient to obtain a high speaker classification accuracy during speech tasks whereas at least eight optimally located sensors are needed to accurately identify these subjects during rest-state (pre-stimuli). These results indicated the unique neural traits of speech production across speakers.