Deep profiling of the speech signal with the use of artificial intelligence in order to increase the security of remote user verification based on the voice

Hubert Rachwalski von Rejchwald

supervisor: Artur Janicki

The research concerns a topic of deep profiling of the speech signal. Main research objectives are the following: i) extraction of a set of features from the audio signal (relevant from the perspective of combating abuse) enabling the desired levels of classification prediction to be obtained; ii) developing a method of transforming source data into features based on experience from the anti-fraud market; iii) creating machine learning models to protect against the widest possible range of adversarial ML attacks in the situation of heavily unbalanced classes; iv) verification of hypotheses of the legitimacy of making the predictive engine independent of linguistic / cultural features as well as the ability to dynamically parametrize the conversation in the context of the diverse needs / specificity of many clients; v) designing a solution that automatically generates scenarios (script) of interaction with the user, supporting the detection of fraud; and vi) verification of hypotheses regarding the parameterization potential of the conversation scenario and the sequence of interactions with a parallel analysis of interactions and immediate issuance of recommendations.

The desired project’s outcome has been visualized below: