Binary audio dataset upgraded in place. Music is label 1; nonmusic is label 0.
The original v21.5 SQLite ledger is restored directly, so existing source
IDs, completed shards, queries and FLEURS locales are not repeated.
Existing speech is untouched. Curated FLEURS/VoxPopuli/MLS/AMI/MLSW/
LibriSpeech labels are trusted. Curated FMA CC0/CC-BY tracks add full-song
genre/artist variety and are sampled for voice-plus-music mixtures.
Mixed, uncertain… See the full description on the dataset page:
https://huggingface.co/datasets/doirty/HaramMusic_Dataset.
