Cross-Lingual Speaker Verification with Domain-Balanced Hard Prototype Mining and Language-Dependent Score Normalization

TL;DR

This paper presents a novel cross-lingual speaker verification method that combines domain-balanced hard prototype mining with language-dependent score normalization, achieving top performance in the SdSV Challenge 2020.

Contribution

It introduces domain-balanced hard prototype mining for effective training and a language-dependent score normalization to improve cross-lingual verification accuracy.

Findings

Achieved a MinDCF of 0.065 and EER of 1.45% on the SdSVC dataset.

Demonstrated the effectiveness of domain-balanced training and language-aware score normalization.

Outperformed previous methods in the SdSV Challenge 2020.

Abstract

In this paper we describe the top-scoring IDLab submission for the text-independent task of the Short-duration Speaker Verification (SdSV) Challenge 2020. The main difficulty of the challenge exists in the large degree of varying phonetic overlap between the potentially cross-lingual trials, along with the limited availability of in-domain DeepMine Farsi training data. We introduce domain-balanced hard prototype mining to fine-tune the state-of-the-art ECAPA-TDNN x-vector based speaker embedding extractor. The sample mining technique efficiently exploits speaker distances between the speaker prototypes of the popular AAM-softmax loss function to construct challenging training batches that are balanced on the domain-level. To enhance the scoring of cross-lingual trials, we propose a language-dependent s-norm score normalization. The imposter cohort only contains data from the Farsi target-domain which simulates the enrollment data always being Farsi. In case a Gaussian-Backend language model detects the test speaker embedding to contain English, a cross-language compensation offset determined on the AAM-softmax speaker prototypes is subtracted from the maximum expected imposter mean score. A fusion of five systems with minor topological tweaks resulted in a final MinDCF and EER of 0.065 and 1.45% respectively on the SdSVC evaluation set.