Misram LLC’s latest published patent application uses machine learning (ML) to determine the probability of spoofing a voice.

With voice authentication becoming more and more common, the probability that a voice has been spoofed, using a synthesized voice or a converted voice, is increasing. As such, many conventional voice-based authentication systems remain highly susceptible to spoofing.

Misram is using a ML, multi-dimensional acoustic feature vector authentication system to build and train multiple multi-dimension acoustic feature vector ML classifiers in order to determine the probability of a voice being spoofed. Their system extracts a number of acoustic features from a user’s voice sample and converts those features into a multi-dimensional acoustic feature vector. Ultimately, the spoofing probability indication is used to determine whether or not to authenticate a user.

This is Misram’s third published document in the AI Cybersecurity sector, classified under the Biometric subcategory.