Abstract:
In bio-related applications privacy is an essential element. While most of the techniques
in Deep Learning rely on single modality, spoofing attacks can be minimized by em ploying multi-modal approaches. Purpose of this research is to develop a technique
in which a person will be given some sentences to speak, audio-visual features will be
merged and using this amalgam of both modalities, language model will validate if the
text read actually validates against the passage given to read. This can be used as an
authentication method to check if the user is actually live and hence can prevent the
print attacks in case of mobile applications.
