Emphasized Non-Target Speaker Knowledge in Knowledge Distillation for Automatic Speaker Verification

TL;DR

This paper improves automatic speaker verification by emphasizing non-target speaker knowledge during knowledge distillation, leading to significant performance gains across multiple models and datasets.

Contribution

It introduces a novel method that highlights non-target speaker probabilities in label-level knowledge distillation, enhancing verification accuracy.

Findings

13.67% average EER reduction on VoxCeleb dataset

Non-target speaker knowledge improves model performance

Method effective across different model architectures

Abstract

Knowledge distillation (KD) is used to enhance automatic speaker verification performance by ensuring consistency between large teacher networks and lightweight student networks at the embedding level or label level. However, the conventional label-level KD overlooks the significant knowledge from non-target speakers, particularly their classification probabilities, which can be crucial for automatic speaker verification. In this paper, we first demonstrate that leveraging a larger number of training non-target speakers improves the performance of automatic speaker verification models. Inspired by this finding about the importance of non-target speakers' knowledge, we modified the conventional label-level KD by disentangling and emphasizing the classification probabilities of non-target speakers during knowledge distillation. The proposed method is applied to three different student model architectures and achieves an average of 13.67% improvement in EER on the VoxCeleb dataset compared to embedding-level and conventional label-level KD methods.