arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2110.05069cs.SDcs.LGeess.AS
收藏

带Patchout的音频Transformers的有效训练

Efficient Training of Audio Transformers with Patchout

Johannes Kepler University Linz(林茨约翰内斯·开普勒大学)

Khaled Koutini, Jan Schlüter, Hamid Eghbal-zadeh, Gerhard Widmer

摘要:

基于transformer的模型在自然语言处理(NLP)中的巨大成功导致了各种尝试,使这些架构适应其他领域,如视觉和音频。最近的研究表明,Transformers在视觉和音频任务上的表现优于卷积神经网络(CNN)。然而,与成熟的CNN相比,Transformer模型的主要缺点之一是计算复杂性。在Transformers中,计算和内存的复杂度与输入长度成二次方增长。因此,在优化Transformers方面进行了大量的工作,但往往以降低预测性能为代价。在这项工作中,我们提出了一种新的方法来优化和正则化音频频谱图上的Transformers。我们提出的模型在Audioset上实现了最先进的性能,并且可以在单个消费级GPU上进行训练。此外,我们提出了一个Transformer模型,它在性能和训练速度方面都优于CNN。源代码:https://github.com/kkoutini/PaSST

英文摘要:

The great success of transformer-based models in natural language processing (NLP) has led to various attempts at adapting these architectures to other domains such as vision and audio. Recent work has shown that transformers can outperform Convolutional Neural Networks (CNNs) on vision and audio tasks. However, one of the main shortcomings of transformer models, compared to the well-established CNNs, is the computational complexity. In transformers, the compute and memory complexity is known to grow quadratically with the input length. Therefore, there has been extensive work on optimizing transformers, but often at the cost of degrading predictive performance. In this work, we propose a novel method to optimize and regularize transformers on audio spectrograms. Our proposed models achieve a new state-of-the-art performance on Audioset and can be trained on a single consumer-grade GPU. Furthermore, we propose a transformer model that outperforms CNNs in terms of both performance and training speed. Source code: https://github.com/kkoutini/PaSST

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。

↑