LSTM Based Speaker Recognition System with Optimized Features
N. M. Nandhitha, S. Emalda Roslin, B. Rajasekhar · 2023
Major challenge in speech processing is due to non-linearity and complexity in speech signals. Characterizing this non-linear signals, affects the performance of the automated speaker recognition system. In this paper, autocorrelation features, Approximation statistical features and detailed statistical features are obtained from the speech signals. These features are fed to RNN and LSTM for speaker recognition. LSTM has higher sensitivity and accuracy than RNN. Similarly, detailed statistical features provide the highest sensitivity when compared to auto correlation features and approximation statistical features.