Why does Self-Supervised Learning for Speech Recognition Benefit Speaker Recognition?
Почему самоконтролируемое обучение для распознавания речи способствует распознаванию говорящего?
2022-09-16
SCID: 54.1/w57qq65s
Discuss with AI
VoxCeleb-1self-supervised learningspeaker recognitionspeaker verificationspeech recognition
Figures from the paper
Abstract (AI)
Recently, self-supervised learning (SSL) has demonstrated strong performance in speaker recognition, even if the pre-training objective is designed for speech recognition. In this paper, we study which factor leads to the success of self-supervised learning on speaker-related tasks, e.g. speaker verification (SV), through a series of carefully designed experiments. Our empirical results on the Voxceleb-1 dataset suggest that the benefit of SSL to SV task is from a combination of mask speech prediction loss, data scale, and model size, while the SSL quantizer has a minor impact. We further employ the integrated gradients attribution method and loss landscape visualization to understand the effectiveness of self-supervised learning for speaker recognition performance.
Key Findings
1
Experiments on VoxCeleb-1 indicate that SSL benefits speaker verification through the combined effects of masked speech prediction loss, training-data scale, and model size.
2
Integrated-gradients attribution and loss-landscape visualization are used to investigate why self-supervised learning improves speaker recognition.
3
Self-supervised learning improves speaker verification despite using pretraining objectives designed for speech recognition.
4
The SSL quantizer has only a minor impact on speaker verification performance compared with masked prediction loss, data scale, and model size.
Research Object
self-supervised speech representations and speaker verification on the VoxCeleb-1 dataset
Research Subject
the contributions of masked speech prediction loss, data scale, model size, and the SSL quantizer to speaker verification performance
Publication Details
Publication Date
2022-09-16
Journal
Publisher
ISSN
Cited by
32
Open access PDF
Access Type
Author Information
Download PDF
Subscribe to digest