Publication Details
An attention-based backend allowing efficient fine-tuning of transformer models for speaker verification
Plchot Oldřich, Ing., Ph.D. (DCGM)
Stafylakis Themos
Mošner Ladislav, Ing. (DCGM)
Burget Lukáš, doc. Ing., Ph.D. (DCGM)
Černocký Jan, prof. Dr. Ing. (DCGM)
Pre-trained model, fine-tuning strategy, speaker verification, attentive pooling
In recent years, self-supervised learning paradigm has received extensive attention due to its great success in various down-stream tasks. However, the fine-tuning strategies for adapting those pre-trained models to speaker verification task have yet to be fully explored. In this paper, we analyze several feature extraction approaches built on top of a pre-trained model, as well as regularization and a learning rate scheduler to stabilize the fine-tuning process and further boost performance: multi-head factorized attentive pooling is proposed to factorize the comparison of speaker representations into multiple phonetic clusters. We regularize towards the parameters of the pretrained model and we set different learning rates for each layer of the pre-trained model during fine-tuning. The experimental results show our method can significantly shorten the training time to 4 hours and achieve SOTA performance: 0.59%, 0.79% and 1.77% EER on Vox1-O, Vox1-E and Vox1-H, respectively.
@inproceedings{BUT185120,
author="Junyi {Peng} and Oldřich {Plchot} and Themos {Stafylakis} and Ladislav {Mošner} and Lukáš {Burget} and Jan {Černocký}",
title="An attention-based backend allowing efficient fine-tuning of transformer models for speaker verification",
booktitle="2022 IEEE Spoken Language Technology Workshop, SLT 2022 - Proceedings",
year="2023",
pages="555--562",
publisher="IEEE Signal Processing Society",
address="Doha",
doi="10.1109/SLT54892.2023.10022775",
isbn="978-1-6654-7189-3",
url="https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10022775"
}