Cosine-Distance Virtual Adversarial Training for Semi-Supervised Speaker-Discriminative Acoustic Embeddings

TL;DR

This paper introduces CD-VAT, a semi-supervised learning method that enhances speaker-discriminative acoustic embeddings by leveraging unlabelled data through cosine-distance based adversarial training, improving speaker verification accuracy.

Contribution

The paper presents a novel semi-supervised training technique, CD-VAT, which does not require unlabelled data to share speaker labels with labelled data, unlike previous methods.

Findings

Achieves 11.1% relative reduction in EER on VoxCeleb dataset.

Demonstrates effectiveness of cosine-distance based adversarial training.

Provides significant improvement over purely supervised baseline.

Abstract

In this paper, we propose a semi-supervised learning (SSL) technique for training deep neural networks (DNNs) to generate speaker-discriminative acoustic embeddings (speaker embeddings). Obtaining large amounts of speaker recognition train-ing data can be difficult for desired target domains, especially under privacy constraints. The proposed technique reduces requirements for labelled data by leveraging unlabelled data. The technique is a variant of virtual adversarial training (VAT) [1] in the form of a loss that is defined as the robustness of the speaker embedding against input perturbations, as measured by the cosine-distance. Thus, we term the technique cosine-distance virtual adversarial training (CD-VAT). In comparison to many existing SSL techniques, the unlabelled data does not have to come from the same set of classes (here speakers) as the labelled data. The effectiveness of CD-VAT is shown on the 2750+ hour VoxCeleb data set, where on a speaker verification task it achieves a reduction in equal error rate (EER) of 11.1% relative to a purely supervised baseline. This is 32.5% of the improvement that would be achieved from supervised training if the speaker labels for the unlabelled data were available.