arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2211.07321cs.CLcs.AIcs.SDeess.AS
收藏

MT4SSL:通过整合多个目标提升自监督语音表示学习

MT4SSL: Boosting Self-Supervised Speech Representation Learning by Integrating Multiple Targets

Shanghai Jiao Tong University(上海交通大学);Tsinghua University(清华大学)

Ziyang Ma, Zhisheng Zheng, Changli Tang, Yujin Wang, Xie Chen

AI总结:

本文提出多任务自监督语音学习框架 MT4SSL,通过结合离线 K-means 目标提取器和在线无梯度教师网络目标提取器,在 LibriSpeech 上以更少数据取得优于或媲美最优模型的性能,并提升预训练收敛效率。

摘要:

在本文中,我们从训练目标如何获取的角度为自监督语音模型提供了一个新的视角。我们将目标提取器概括为离线目标提取器(Off-TE)和在线目标提取器(On-TE)。基于此,我们提出了一种新的用于自监督学习的多任务学习框架 MT4SSL,即通过整合多个目标提升自监督语音表示学习。MT4SSL 分别使用 K-means 算法作为 Off-TE,并使用一个无梯度的教师网络作为 On-TE。我们的模型在 LibriSpeech 基准上以非平凡的差距优于先前的自监督学习方法,并且在数据更少的情况下与表现最好的模型相当甚至更好。此外,我们发现同时使用 Off-TE 和 On-TE 能在预训练阶段获得更好的收敛性。鉴于其有效性和高效性,我们认为从我们的视角对自监督语音模型进行多任务学习是一个有前景的趋势。

英文摘要:

In this paper, we provide a new perspective on self-supervised speech models from how the training targets are obtained. We generalize the targets extractor into Offline Targets Extractor (Off-TE) and Online Targets Extractor (On-TE). Based on this, we propose a new multi-tasking learning framework for self-supervised learning, MT4SSL, which stands for Boosting Self-Supervised Speech Representation Learning by Integrating Multiple Targets. MT4SSL uses the K-means algorithm as an Off-TE and a teacher network without gradients as an On-TE, respectively. Our model outperforms previous SSL methods by nontrivial margins on the LibriSpeech benchmark, and is comparable to or even better than the best-performing models with fewer data. Furthermore, we find that using both Off-TE and On-TE results in better convergence in the pre-training phase. With both effectiveness and efficiency, we think doing multi-task learning on self-supervised speech models from our perspective is a promising trend.

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。

↑