ACAVCaps

dataset
Xiaomi
38,000-hour collection of general audio captions for training holistic audio-reasoning models.
Paper (arXiv)GitHub
audiotraining-data