Speech/Non-Speech Dataset for Training Speech Detection Models (5-second clips)

This dataset contains a collection of .wav audio clips from Freesound.org, and a corresponding metadata.csv file. Building on the previous dataset: https://doi.org/10.5281/zenodo.15731210, this dataset is made up of 5-second segments for easier model training. To avoid data leakage, these clips were segmented after the train/val/test split, which is included in the metadata file. The final dataset breakdown is:

- 1802 training clips: 855 speech, 947 non-speech- 256 validation clips: 128 speech, 128 non-speech- 219 test clips: 102 speech, 117 non-speech