Speaker Recognition Based on Mutated Monarch Butterfly Optimization Configured Artificial Neural Network

Abstract

Speaker recognition is the process of extracting speaker-specific details from voice waves to validate the features asserted by system users; in other words, it allows voice-controlled access to a range of services. The research initiates with extraction features from voice signals and employing those features in Artificial Neural Network (ANN) for speaker recognition. Increasing the number of hidden layers and their associated neurons reduces the training error and increases the computational process's complexity. It is essential to have an optimal number of hidden layers and their corresponding, but attaining those optimal configurations through a manual or trial and the process takes time and makes the process more complex. This urges incorporating optimization approaches for finding optimal hidden layers and their corresponding neurons. The technique involve in configuring the ANN is Mutated Monarch Butterfly Optimization (MMBO). The proposed MMBO employed for configuring the ANN achieves the sensitivity of 97.5% in a real- time database that is superior to contest techniques.