Machine Learning and Metaheuristic Algorithms for Voice-Based Authentication: A Mobile Banking Case Study

These days, every bank requires secure identification before granting access to personal accounts. Voice authentication is increasingly popular for important mobile processes. The hybrid authentication technique is used in this work to address concerns about forgery attacks in the voice authentication system. Here, we accomplish this goal by using a user authentication approach that decodes passwords from acoustic signals. This method involves the user speaking into the phone to enter their private password. Then, to crack the code, an artificial neural network is employed along with retrieved statistics and speech parameters such as energy and Mel frequency cepstral coefficient. The password numbers are read and saved. The planned application divides the read integers by using the pauses in between the digits. On the other hand, by employing the frequency and phase properties to compare the target's voice to the discovered password key, this method will prevent speech forgeries. This speech and password matching verification system uses our fuzzy nonlinear support vector machine network classification system, which was trained using the Ali Baba and the forty thieves algorithm. Our method is evaluated on a dataset of 30 individuals and three smartphones, achieving an accuracy rate of more than 98.29%. Our system is resistant against a wide range of challenges, including variations in authentication angle, authentication distance, passphrase length, ambient noise, and more, in addition to being device independent.