Skip to content

audio folder builder cannot detect custom split name #7529

Open
@phineas-pta

Description

@phineas-pta

Describe the bug

when using audio folder builder (load_dataset("audiofolder", data_dir="/path/to/folder")), it cannot detect custom split name other than train/validation/test

Steps to reproduce the bug

i have the following folder structure

my_dataset/
├── train/
│   ├── lorem.wav
│   ├── …
│   └── metadata.csv
├── test/
│   ├── ipsum.wav
│   ├── …
│   └── metadata.csv
├── validation/
│   ├── dolor.wav
│   ├── …
│   └── metadata.csv
└── custom/
    ├── sit.wav
    ├── …
    └── metadata.csv

using ds = load_dataset("audiofolder", data_dir="/path/to/my_dataset")

Expected behavior

i got ds with only 3 splits train/validation/test, whenever i rename train/validation/test folder it also disappear if i re-create ds

Environment info

  • datasets version: 3.5.0
  • Platform: Windows-11-10.0.26100-SP0
  • Python version: 3.12.8
  • huggingface_hub version: 0.30.2
  • PyArrow version: 18.1.0
  • Pandas version: 2.2.3
  • fsspec version: 2024.9.0

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