# ERes2NetV2: Boosting Short-Duration Speaker Verification Performance with Computational Efficiency

Source: [ERes2NetV2: Boosting Short-Duration Speaker Verification Performance with Computational Efficiency](<https://www.sota2.com/research/paper/68fa80141e209131cea9be93>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2025-10-27T07:13:29.356Z

Authors: Yafeng Chen, Siqi Zheng, Hui Wang, Luyao Cheng, Qian Chen, Shiliang Zhang, Junjie Li

Published: 2024-06-04T09:56:39.000Z

## Abstract

Speaker verification systems experience significant performance degradation when tasked with short-duration trial recordings. To address this challenge, a multi-scale feature fusion approach has been proposed to effectively capture speaker characteristics from short utterances. Constrained by the model's size, a robust backbone Enhanced Res2Net (ERes2Net) combining global and local feature fusion demonstrates sub-optimal performance in short-duration speaker verification. To further improve the short-duration feature extraction capability of ERes2Net, we expand the channel dimension within each stage. However, this modification also increases the number of model parameters and computational complexity. To alleviate this problem, we propose an improved ERes2NetV2 by pruning redundant structures, ultimately reducing both the model parameters and its computational cost. A range of experiments conducted on the VoxCeleb datasets exhibits the superiority of ERes2NetV2, which achieves EER of 0.61% for the full-duration trial, 0.98% for the 3s-duration trial, and 1.48% for the 2s-duration trial on VoxCeleb1-O, respectively.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Speaker Identification on Vox1-O](<https://www.sota2.com/research/sota/speaker-identification-on-vox1-o.md>): Speaker Identification; Vox1-O; EER: 1.28.

- [Speaker Recognition on VoxCeleb1 original (vox1-o)](<https://www.sota2.com/research/sota/speaker-recognition-on-voxceleb1-original-vox1-o.md>): Speaker Recognition; VoxCeleb1 original (vox1-o); EER (mean): 1.72.

- [Open-set Speaker Identification on VoxCeleb2 (test)](<https://www.sota2.com/research/sota/open-set-speaker-identification-on-voxceleb2-test.md>): Open-set speaker identification; VoxCeleb2 (test); EER: 3.74.

- [Open-set speaker identification on ESD (test)](<https://www.sota2.com/research/sota/open-set-speaker-identification-on-esd-test.md>): Open-set speaker identification; ESD (test); EER: 4.28.

- [Open-set speaker identification on 3D-Speaker (test)](<https://www.sota2.com/research/sota/open-set-speaker-identification-on-3d-speaker-test.md>): Open-set speaker identification; 3D-Speaker (test); EER: 3.59.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2406.02167>)

[PDF](<https://arxiv.org/pdf/2406.02167.pdf>)
