Combined Speaker and Speech Ascent Recognition Aided Authentication Using Adaptive Convolutional Neural Network with Attention Mechanism

Abstract

Speaker recognition is a multimodal sensory technology that identifies people by their voices. Considering the safety and important usage of these types of systems, improved performance is critical (for example, voice-based transactions, control of access, and criminal detection). As voice recognition systems become integrated into future electronic devices, recognition of speakers will serve as an organic authentication method. Despite the fact they have already been available for over forty years, they aren't generally accepted as independent security measures due to their unacceptable inadequate performance, i.e., large false recognition and denial rates. In this paper, the efficient speaker and speech ascent recognition model is developed to provide automated speaker authentication and speech detection. At first, the desired input speech signals are collected from the benchmark resources. Subsequently, the spectral features include flux, spectral flatness, spectral bandwidth, spectral Centroid, and spectral contrast, similarly, the cepstral features like Linear Predictive Coding Coefficient (LPCC), Mel Frequency Cepstral Coefficient (MFCC), and the deep features are extracted from the input signal using the autoencoder structure. After the feature extraction process, the new structure called Adaptive Convolutional Neural Network with Attention mechanism (ACNN-AM) is developed to perform the speaker and speech recognition process. The ACNN-AM structure initially performs the speaker recognition process, which is considered the most suitable method to verify the authenticity of the user in the smart phone environment. After the speaker's recognition, the authenticity of the speech is verified by the same ACNN-AM. The access is provided to the user if the speaker and speech are matched. The speaker and speech recognition performance of the ACNN-AM is enhanced by optimizing the attributes using Enhanced Mountaineering Team-Based Optimization (EMTBO). Finally, the experimental analysis is conducted to prove the efficacy of the developed model.