The effects of distance and reverberation time on speaker recognition performance
摘要
This study investigates speaker recognition systems designed for distant speakers, considering challenges related to microphone placement variations, background noise, and reverberation. The key factors affecting reverberation are the reverberation time (RT) and the distance between the speaker and the microphones. The study utilizes Mel-frequency cepstral coefficients (MFCC) and assesses their interplay with RT and distance, measuring verification outcomes through the equal error rate (EER). Findings show that, at a 0.53-s RT and a 0.5-cm distance, the EER is 2.33%, but at a 7-m distance, it rises to 6%. A longer 1.5-s RT and a 5-m distance result in a notable EER increase of 14.66%. The system exhibits resilience to minor distance variations, excelling in close distances and short reverberation times, with optimal accuracy when the microphone-to-source distance is less than 0.5 m.