some questions when reproducing your results

[ben100118]

Hello Yuan,I'm delighted to read your paper and reproduce your work.And I encounter some problems.
When empolying the audioset_pretrain,why does the stride as same as the patch_size(overlap == 0).
In addition, when I train ast on speechCommands,the acc is lower than your result（value 0.1）
Thank you for taking the time to read my question; I look forward to your response！

[ben100118]

Furthermore，what's the value of mixup?

[YuanGongND]

0.1 must be some bug.

Please check the reference of https://arxiv.org/abs/2102.01243 this paper (i.e., search mixup and find the related paper). This is a data augmentation method.