# Analyzing speaker verification embedding extractors and back-ends under language and channel mismatch

arXiv: https://arxiv.org/abs/2203.10300 · HTML: https://mlsanity.com/paper/2203.10300/

## Key facts

- Cited by 5 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 19 Mar 2022.

## Built on by (most cited first, 5 of 5)

- [Probabilistic Spherical Discriminant Analysis: An Alternative to PLDA for length-normalized embeddings](https://mlsanity.com/paper/2203.14893/) (2022)
- [Universal speaker recognition encoders for different speech segments duration](https://mlsanity.com/paper/2210.16231/) (2022)
- [Toroidal Probabilistic Spherical Discriminant Analysis](https://mlsanity.com/paper/2210.15441/) (2022)
- [Post-Training Embedding Alignment for Decoupling Enrollment and Runtime Speaker Recognition Models](https://mlsanity.com/paper/2401.12440/) (2024)
- [The Voiceprint Fallacy: Why Voices Are Not Unique Biometric Imprints](https://mlsanity.com/paper/2608.07980/) (2026)

## Abstract (arXiv)

In this paper, we analyze the behavior and performance of speaker embeddings and the back-end scoring model under domain and language mismatch. We present our findings regarding ResNet-based speaker embedding architectures and show that reduced temporal stride yields improved performance. We then consider a PLDA back-end and show how a combination of small speaker subspace, language-dependent PLDA mixture, and nuisance-attribute projection can have a drastic impact on the performance of the system. Besides, we present an efficient way of scoring and fusing class posterior logit vectors recently shown to perform well for speaker verification task. The experiments are performed using the NIST SRE 2021 setup.