Formants and Prosodic Features' Effects on Arabic Speaker Identification Accuracy in Noisy Environments

This study investigates the use of formants and prosodic features, specifically pitch and intensity, for speaker identification in realconditions. To enhance the robustness of the acoustic models against speech signal variations in noisy environments, Mel-Frequency Cepstral Coefficient (MFCC) are added to these features. A Speaker Identification system based on Hidden Markov Models (HMM) is implemented in the independent text mode. The combination of formants and prosodic features with cepstral features improves the identification accuracy, particularly in high-noise environments, up to 10%, in comparison to an MFCC based system. The results show that the use of multivariate feature vectors significantly improves the performance of an identification system in the presence of noise compared to an MFCC-based system.