AudioSet v2: 48 kHz audio, per-clip provenance, measurements, duplicate flags and missing-segment reasons; first download kept under audio_v1/

Dataset card, Google label files, per-clip metadata with the shard map, noise-bank subset