Spurious label column created when audiofolder/imagefolder directories match split names

[neha222222]

Describe the bug

When using audiofolder or imagefolder with directories for splits (train/test) rather than class labels, a spurious label column is incorrectly created.

Example: https://huggingface.co/datasets/datasets-examples/doc-audio-4

from datasets import load_dataset
ds = load_dataset("datasets-examples/doc-audio-4")
print(ds["train"].features)

Shows 'label' column with ClassLabel(names=['test', 'train']) - incorrect!## Root cause

In folder_based_builder.py, the labels set is accumulated across ALL splits (line 77). When directories are train/ and test/:

• labels = {"train", "test"} → len(labels) > 1 → add_labels = True
• Spurious label column is created with split names as class labels

Expected behavior

No label column should be added when directory names match split names.

Proposed fix

Skip label inference when inferred labels match split names.

cc @lhoestq