SoundingEarth

SoundingEarth SoundingEarth is a geo-referenced soundscape dataset that pairs Google Earth imagery with geotagged environmental audio recordings originally collected by Heidler et al. for self-supervised audiovisual representation learning. Splits Split Rows train 40,241 val 3,242 test 5,801 Train/val/test splits were created using a cell-based geographic partitioning strategy introduced in the PSM paper — the Earth’s surface is divided into spatial… See the full description on the dataset page:

Source: Hugging Face Hub (MVRL/SoundingEarth). Metadata imported from the dataset’s Hub tags.