Design on Small-Footprint x-Vector-Based End-to-End Speaker Verification System

The uniqueness of a person's voice has been increasingly used for identity recognition in recent years. Most of the research focuses on improving accuracy with unlimited computational resources, and few studies focus on the implementation of the overall system and testing it in real-world conditions. In this paper, we design an end-to-end speaker verification (SV) system with an x-vector neural network. Since the x-vector is one of the most robust neural network models for speaker verification, it can achieve good accuracy in some complex recognition environments. To maximize computational efficiency while maintaining recognition accuracy, we simplified the architecture of the x-vector and employed pruning to remove unnecessary weights. Additionally, we designed a Graphical User Interface (GUI) to make it easier for the user to operate our system. The proposed system can achieve real-time recognition speed and high accuracy in real-world testing.