Using wavelet scattering transform to create voiceprint of a password

Abstract

Today, new biometric technologies are increasingly being used in various protocols and interfaces that implement user identification and verification. Voice identification, which implements text-dependent and text-independent speech recognition, is widely exploited in the human-machine interface. An example is the ID R&D developer, owned by the Mitek group of companies, which offers an AI-based speaker recognition product IDVoice that combines three-modal biometric capture with liveness detection, digital ID issuance, and mobile authentication. The developed SDK of ID R&D produces so-called a “voiceprint” that is a template analogous to someone’s fingerprint and capable to perform user verification. Usually Shallow and Deep Neural Networks (DNN) are used in these technologies.