EEG and voice bimodal biometric authentication scheme with fusion at signal level

Biometric systems, as means of identity recognition, are fundamental for the defense and protection of human rights, hence the importance of having systems that, in addition to presenting high performance, meet acceptable levels of universality, uniqueness and circumvention among other characteristics. This paper presents two single-modality recognition schemes and a bimodal user identification system. The single-modality experiments consist of a voice-based identity verification system using Hidden Markov Models as classifier involving 51 users, and an EEG-based user identification system, employing an artificial neuronal network, for 50 users. The bimodal trial fuses voice and EEG modalities using a mixed signal-level-decision-level fusion scheme, with Hidden Markov Models on the first classification stage and majority vote for the final classification result, considering 50 users. An accuracy of 83.43% is achieved for a set of 7-state HMM's with an 80% voice-20% EEG mixed signal. These results comply with our expectations, considering the fact that the experiment is designed with short-length (2 seconds) samples.