Cross attentive pooling for speaker verification
This addresses speaker verification for noisy video data, offering an incremental improvement over existing pooling methods.
The paper tackles text-independent speaker verification in noisy 'in the wild' videos by proposing Cross Attentive Pooling (CAP), which uses context across reference-query pairs to generate discriminative embeddings, outperforming comparable pooling strategies on the VoxCeleb dataset.
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.