Speaker verification for telemedical applications

A major problem in telemedical applications is the confidentiality and secured access to patient/medical records. The purpose of this investigation was to develop neural network models for speech-based speaker verification. Speech data was analyzed and fed to several neural networks. Based on the performance of the initial testing, the best five networks were recruited into the committee. The majority opinion of the committee was used to classify the designated individual and the imposters. The committee correctly verified the designated individual in 100% of the cases and rejected the imposters in 100% of the cases. The committee networks-based user verification has a significant potential for telemedical applications