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 2406.04269eess.AScs.SD
收藏

超越性能平台期:语音增强可扩展性的全面研究

Beyond Performance Plateaus: A Comprehensive Study on Scalability in Speech Enhancement

Shanghai Jiao Tong University(上海交通大学);Carnegie Mellon University(卡内基梅隆大学);Waseda University(早稻田大学)

Wangyou Zhang, Kohei Saijo, Jee-weon Jung, Chenda Li, Shinji Watanabe, Yanmin Qian

AI总结:

本文全面研究了语音增强模型在架构、模型规模、计算预算和数据集规模上的可扩展性,发现其缩放效应与语音识别等任务的异同,为构建大规模多领域语料库和高效可扩展架构提供了新见解。

摘要:

基于深度学习的语音增强(SE)模型在过去十年中取得了令人瞩目的性能。人们设计了众多先进架构以提供最先进的性能;然而,它们的可扩展性潜力仍未被揭示。同时,大多数研究集中于规模受限、多样性有限的小型数据集,导致性能提升出现平台期。在本文中,我们旨在通过从架构、模型规模、计算预算和数据集规模四个方面探索SE模型的可扩展性,为解决上述问题提供新见解。我们的研究涉及几种流行的SE架构和来自不同领域的语音数据。实验揭示了SE中的缩放效应与语音识别等其他任务之间的相似性与差异。这些发现进一步为探索不足的SE方向提供了见解,例如更大规模的多领域语料库和高效可扩展的架构。

英文摘要:

Deep learning-based speech enhancement (SE) models have achieved impressive performance in the past decade. Numerous advanced architectures have been designed to deliver state-of-the-art performance; however, their scalability potential remains unrevealed. Meanwhile, the majority of research focuses on small-sized datasets with restricted diversity, leading to a plateau in performance improvement. In this paper, we aim to provide new insights for addressing the above issues by exploring the scalability of SE models in terms of architectures, model sizes, compute budgets, and dataset sizes. Our investigation involves several popular SE architectures and speech data from different domains. Experiments reveal both similarities and distinctions between the scaling effects in SE and other tasks such as speech recognition. These findings further provide insights into the under-explored SE directions, e.g., larger-scale multi-domain corpora and efficiently scalable architectures.

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。

↑