Back-ends Selection for Deep Speaker Embeddings

TL;DR

This paper systematically compares cosine similarity and PLDA back-ends for deep speaker embeddings, finding that cosine excels in same-domain scenarios while PLDA is preferable for cross-domain situations, based on extensive experiments.

Contribution

It provides a comprehensive analysis and practical guidelines for selecting back-ends for deep speaker embeddings in various domain conditions.

Findings

Cosine similarity outperforms PLDA in same-domain scenarios.

PLDA yields better results in cross-domain situations.

Experimental validation on VoxCeleb and NIST SRE datasets supports the conjecture.

Abstract

Probabilistic Linear Discriminant Analysis (PLDA) was the dominant and necessary back-end for early speaker recognition approaches, like i-vector and x-vector. However, with the development of neural networks and margin-based loss functions, we can obtain deep speaker embeddings (DSEs), which have advantages of increased inter-class separation and smaller intra-class distances. In this case, PLDA seems unnecessary or even counterproductive for the discriminative embeddings, and cosine similarity scoring (Cos) achieves better performance than PLDA in some situations. Motivated by this, in this paper, we systematically explore how to select back-ends (Cos or PLDA) for deep speaker embeddings to achieve better performance in different situations. By analyzing PLDA and the properties of DSEs extracted from models with different numbers of segment-level layers, we make the conjecture that Cos is better in same-domain situations and PLDA is better in cross-domain situations. We conduct experiments on VoxCeleb and NIST SRE datasets in four application situations, single-/multi-domain training and same-/cross-domain test, to validate our conjecture and briefly explain why back-ends adaption algorithms work.