Speaker Verification Using Embedding Parameters in Noisy Environments (Transport Vehicles)

This research paper presents a speaker verification system employing speaker embedding parameters, specifically I-vectors derived from a Gaussian Mixture Model-Universal Background Model (GMM-UBM), utilizing speech samples collected from diverse transport vehicles such as cars, buses, and trains. Mel Frequency Cepstral Coefficients (MFCCs) serve as low-level features for this system, and experiments were conducted in both clean and noisy environments. The results show a notable decrease in verification accuracy as noise levels increase. To address this, a noise-robust low-level feature, autocorrelated MFCC (AMFCC), is introduced, revealing its adaptability to noisy conditions. The I-vectors derived from AMFCC significantly improve recognition accuracy, highlighting the potential for enhancing the robustness of speaker verification systems in challenging acoustic environments, particularly in the context of transport vehicles.