Improving Speaker Verification with Self-Pretrained Transformer Models

Improving Speaker Verification with Self-Pretrained Transformer Models

Paper in proceedings (conference paper)

Recently, fine-tuning large pre-trained Transformer models using downstream datasets has received a rising interest. Despite their success, it is still challenging to disentangle the benefits of large-scale datasets and Transformer structures from the limitations of the pre-training. In this paper, we introduce a hierarchical training approach, named self-pretraining, in which Transformer models are pretrained and finetuned on the same dataset. Three pre-trained models including HuBERT, Conformer andWavLM are evaluated on four different speaker verification datasets with varying sizes. Our experiments show that these self-pretrained models achieve competitive performance on downstream speaker verification tasks with only one-third of the data compared to Librispeech pretraining, such as Vox- Celeb1 and CNCeleb1. Furthermore, when pre-training only on the VoxCeleb2-dev, the Conformer model outperforms the one pre-trained on 94k hours of data using the same fine-tuning settings.

Recently, fine-tuning large pre-trained Transformer models using downstream datasets has received a rising interest. Despite their success, it is still challenging to disentangle the benefits of large-scale datasets and Transformer structures from the limitations of the pre-training. In this paper, we introduce a hierarchical training approach, named self-pretraining, in which Transformer models are pretrained and finetuned on the same dataset. Three pre-trained models including HuBERT, Conformer andWavLM are evaluated on four different speaker verification datasets with varying sizes. Our experiments show that these self-pretrained models achieve competitive performance on downstream speaker verification tasks with only one-third of the data compared to Librispeech pretraining, such as Vox- Celeb1 and CNCeleb1. Furthermore, when pre-training only on the VoxCeleb2-dev, the Conformer model outperforms the one pre-trained on 94k hours of data using the same fine-tuning settings.

speaker verification, pre-trained speech transformer model, pre-training,

speaker verification, pre-trained speech transformer model, pre-training,

PENG, J.; PLCHOT, O.; STAFYLAKIS, T.; MOŠNER, L.; BURGET, L.; ČERNOCKÝ, J.

2024

20.08.2023

International Speech Communication Association

Dublin

Proceedings of the Annual Conference of the International Speech Communication Association, INTERSPEECH

1990-9772

Proceedings of Interspeech

2023

08

French Republic

5361

5365

5

https://www.isca-speech.org/archive/pdfs/interspeech_2023/peng23_interspeech.pdf

peng23_interspeech2023_improving