Cross attentive pooling for speaker verification

TL;DR

This paper introduces Cross Attentive Pooling (CAP), a novel approach for text-independent speaker verification that leverages pair-wise context to produce more discriminative utterance embeddings, outperforming existing pooling methods on VoxCeleb.

Contribution

The paper proposes a new cross attentive pooling method that incorporates pair-wise context for improved speaker verification performance.

Findings

CAP outperforms existing pooling strategies on VoxCeleb.

Utilizes pair-wise context to enhance embedding discriminability.

Improves accuracy in real-world 'in the wild' scenarios.

Abstract

The goal of this paper is text-independent speaker verification where utterances come from 'in the wild' videos and may contain irrelevant signal. While speaker verification is naturally a pair-wise problem, existing methods to produce the speaker embeddings are instance-wise. In this paper, we propose Cross Attentive Pooling (CAP) that utilizes the context information across the reference-query pair to generate utterance-level embeddings that contain the most discriminative information for the pair-wise matching problem. Experiments are performed on the VoxCeleb dataset in which our method outperforms comparable pooling strategies.