Source -Free Domain Adaptation for Speaker Verification in Data-Scarce Languages and Noisy Channels

TL;DR

This paper proposes source-free domain adaptation techniques for speaker verification in data-scarce languages and noisy channels, addressing privacy and resource limitations by exploring fine-tuning and a new cluster-learn algorithm.

Contribution

It introduces a novel iterative cluster-learn algorithm and evaluates fine-tuning methods for source-free adaptation in challenging speech verification scenarios.

Findings

Fine-tuning improves performance with limited target data.

The cluster-learn algorithm effectively adapts to unlabeled target data.

Methods outperform baseline approaches in noisy and language-mismatched conditions.

Abstract

Domain adaptation is often hampered by exceedingly small target datasets and inaccessible source data. These conditions are prevalent in speech verification, where privacy policies and/or languages with scarce speech resources limit the availability of sufficient data. This paper explored techniques of sourcefree domain adaptation unto a limited target speech dataset for speaker verificationin data-scarce languages. Both language and channel mis-match between source and target were investigated. Fine-tuning methods were evaluated and compared across different sizes of labeled target data. A novel iterative cluster-learn algorithm was studied for unlabeled target datasets.