A method is provided that includes configuring a cellular phone to function as a mobile debit card by installing an application for contactless interfacing with ATMs. The method supplements the application with a neural network (NN) based biometric verification process configured to reduce an incorrect user error value over time to increasingly harden the application to undesired intrusion. The NN based biometric verification process comprises: performing an initial face recognition; greeting the user with one of a plurality of questions in a specific language of the user and evaluating a pre-defined answer provided from the user in the specific language; and detecting, in an acoustic utterance having dialogue in support of an ATM session, a voice and a prosody style indicative of the user in combination with lip and face movements made by the user corresponding to and in synchronization with the acoustic utterance.