Learning Metrics from Mean Teacher: A Supervised Learning Method for Improving the Generalization of Speaker Verification System

TL;DR

This paper introduces a supervised learning method using the Mean Teacher model to enhance speaker verification systems' ability to generalize to unseen speakers by producing more stable and discriminative speaker embeddings.

Contribution

The paper applies the Mean Teacher approach to speaker verification, demonstrating improved generalization and embedding stability over traditional methods.

Findings

11.61% relative performance improvement on VoxCeleb1

Mean Teacher produces more accurate speaker embeddings

Enhanced discrimination between speakers

Abstract

Most speaker verification tasks are studied as an open-set evaluation scenario considering the real-world condition. Thus, the generalization power to unseen speakers is of paramount important to the performance of the speaker verification system. We propose to apply \textit {Mean Teacher}, a temporal averaging model, to extract speaker embeddings with small intra-class variance and large inter-class variance. The mean teacher network refers to the temporal averaging of deep neural network parameters; it can produces more accurate and stable representations than using weights after the training finished. By learning the reliable intermediate representation of the mean teacher network, we expect that the proposed method can explore more discriminatory embedding spaces and improve the generalization performance of the speaker verification system. Experimental results on the VoxCeleb1 test set demonstrate that the proposed method relatively improves performance by 11.61\%, compared to a baseline system.