Graph Attention Networks for Speaker Verification

TL;DR

This paper introduces a graph attention network-based framework for speaker verification that models segment embeddings as graph nodes, achieving significant accuracy improvements over traditional methods.

Contribution

The novel use of graph attention networks to interpret segment-wise speaker embeddings as graphs for improved speaker verification accuracy.

Findings

Achieved an average 20% reduction in equal error rate over cosine similarity baseline.

Validated effectiveness across three different speaker embedding extractors.

Demonstrated consistent performance improvements with the proposed framework.

Abstract

This work presents a novel back-end framework for speaker verification using graph attention networks. Segment-wise speaker embeddings extracted from multiple crops within an utterance are interpreted as node representations of a graph. The proposed framework inputs segment-wise speaker embeddings from an enrollment and a test utterance and directly outputs a similarity score. We first construct a graph using segment-wise speaker embeddings and then input these to graph attention networks. After a few graph attention layers with residual connections, each node is projected into a one-dimensional space using affine transform, followed by a readout operation resulting in a scalar similarity score. To enable successful adaptation for speaker verification, we propose techniques such as separating trainable weights for attention map calculations between segment-wise speaker embeddings from different utterances. The effectiveness of the proposed framework is validated using three different speaker embedding extractors trained with different architectures and objective functions. Experimental results demonstrate consistent improvement over various baseline back-end classifiers, with an average equal error rate improvement of 20% over the cosine similarity back-end without test time augmentation.