Reverberation Time and Distance Impact on the Equal Error Rate

Recently, speaker recognition has emerged as a pivotal biometric tool for finding security, forensics, and banking applications. Despite its widespread adoption, real-world challenges, such as variations in environmental conditions like reverberation time and background noise, pose significant obstacles to speaker recognition. Among the well-established methods for speaker recognition, MFCC and GFCC are notable for their maturity and widespread use. The study delves into the impact of reverberation conditions, characterized by parameters like reverberation time and direct-to-reverberation ratio, on the performance of speaker verification systems. The investigation includes an analysis of the relationship between these parameters and verification results, measured through the equal error rate percentage. Regression models are introduced to elucidate the correlation between the equal error rate, reverberation time, and the distance from the microphone source. The results emphasize the optimal microphone accuracy when positioned within 0.5 m of the sound source. Furthermore, the study highlights the superior performance of GFCC compared to MFCC in the conducted evaluations.