arXiv 2308.027742024-12-27eess.AScs.SD

自蒸馏原型网络：在无监督条件下学习鲁棒的说话人表示

Self-Distillation Prototypes Network: Learning Robust Speaker Representations without Supervision

Alibaba Group（阿里巴巴集团）

Yafeng Chen, Siqi Zheng, Hui Wang, Luyao Cheng, Qian Chen, Chong Deng, Shiliang Zhang, Wen Wang

备注：: arXiv admin note: text overlap with arXiv:2211.04168

阅读 PDF

arXiv 页面 ↗HTML 原文 ↗

AI总结：

本文提出自蒸馏原型网络 SDPN，通过将增强视图与原视图对齐到相同可学习原型并引入多样性正则化防止模型坍缩，在无说话人标签条件下于 VoxCeleb1 上取得最先进的说话人验证性能。

摘要：

在没有显式说话人标签的情况下训练具有说话人区分性和鲁棒性的说话人验证系统仍然是一个持续的挑战。在本文中，我们提出了一种新颖的自监督说话人验证方法，即自蒸馏原型网络（Self-Distillation Prototypes Network，SDPN），它有效地促进了自监督说话人表示学习。SDPN 将话语增强视图的表示分配到与原始视图表示相同的原型上，从而在增强视图和原始视图之间实现有效的知识迁移。由于 SDPN 训练过程中缺乏负样本对，网络倾向于在嵌入空间中将正样本对对齐得过于紧密，这种现象被称为模型坍缩。为了缓解这一问题，我们在 SDPN 的嵌入中引入了一个多样性正则化项。在 VoxCeleb 数据集上的综合实验证明了 SDPN 在自监督说话人验证方法中的优越性。SDPN 在 VoxCeleb1 说话人验证评估基准上创造了新的最先进水平，在训练中不使用任何说话人标签的情况下，在 VoxCeleb1-O、VoxCeleb1-E 和 VoxCeleb1-H 三种测试条件下分别取得了 1.80%、1.99% 和 3.62% 的等错误率（Equal Error Rate）。消融研究表明，所提出的自蒸馏网络中的可学习原型和多样性正则化都对验证性能有所贡献。

英文摘要：

Training speaker-discriminative and robust speaker verification systems without explicit speaker labels remains a persistent challenge. In this paper, we propose a novel self-supervised speaker verification approach, Self-Distillation Prototypes Network (SDPN), which effectively facilitates self-supervised speaker representation learning. SDPN assigns the representation of the augmented views of an utterance to the same prototypes as the representation of the original view, thereby enabling effective knowledge transfer between the augmented and original views. Due to lack of negative pairs in the SDPN training process, the network tends to align positive pairs quite closely in the embedding space, a phenomenon known as model collapse. To mitigate this problem, we introduce a diversity regularization term to embeddings in SDPN. Comprehensive experiments on the VoxCeleb datasets demonstrate the superiority of SDPN among self-supervised speaker verification approaches. SDPN sets a new state-of-the-art on the VoxCeleb1 speaker verification evaluation benchmark, achieving Equal Error Rate 1.80%, 1.99%, and 3.62% for trial VoxCeleb1-O, VoxCeleb1-E and VoxCeleb1-H, without using any speaker labels in training. Ablation studies show that both proposed learnable prototypes in self-distillation network and diversity regularization contribute to the verification performance.

机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。