Research on Speaker Recognition Based on the ECAPA-TDNN Method
Zhenye Gan, Xu Zhang · 2024
Speaker recognition, also known as speaker identification, encompasses two major functions: speaker identification and speaker verification. It is a critical branch of speech processing. The aim of this technology is to recognize or verify the identity of a speaker. Speaker recognition technologies are widely applied in various domains, including security authentication, law enforcement, and personalized interaction systems. Despite their extensive application, speaker recognition technologies face numerous challenges in practical applications such as noise interference and recording quality. Given the limited research on speaker recognition for ethnic minorities within the country, this study employs an improved ECAPATDNN model and conducts experimental validation on a selfconstructed Tibetan speaker recognition dataset. The experimental results indicate an Equal Error Rate (EER) of 0.013, a True Positive Rate (TPR) of 0.992, and a False Positive Rate (FPR) of 0.006, with the threshold set at 0.30.