Deep Learning for Speaker Recognition: A Comparative Analysis of 1D-CNN and LSTM Models Using Diverse Datasets

Speaker recognition is a vital component of identity verification and security systems that has made significant progress through the use of deep neural networks. This article examines the comparative performance of two neural network models, namely a one-dimensional convolutional neural network (1D-CNN) and a long short-term memory (LSTM-based) network in identifying individuals based on their recorded voices. The article uses three diverse datasets, including the Raparin Artificial Intelligent Lab (RAIL) dataset, which was created locally, and two public datasets, namely the TIMIT dataset and the Jordan dataset. The results show that the proposed 1D-CNN model consistently outperforms the LSTM model and has a significant accuracy rate, especially in the RAIL dataset, which achieves an accuracy of over 95.22%. This study emphasizes the potential of deep learning algorithms in improving sound-based identity recognition and highlights its effects on system security and communications.