Skip to content

Low Accuracy on ESC-50 #3

Description

@The-Mats

Hi,
I cloned the repisotory and performed all the preparation steps using a huggingface hosted Audioset (~2m Samples):
https://huggingface.co/datasets/confit/audioset-full

When performing a training I am only getting ESC-50 results of roughly 60%! Why is that and has anyone else managed to replicate the performance from the paper?
Are the default pipeline settings the same as the paper?

Thanks in advance :)

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions