OpenMIC-2018 (zenodo) is a dataset for polyphonic instruments identification.
Use openmic2008/prepare_scripts/download_preprocess.py to download and pack the dataset:
cd openmic2008/prepare_scripts/
python download_preprocess.pyWhen the script completes, you should have two files inside audioset_hdf5s/mp3/
openmic_train.csv_mp3.hdf and openmic_test.csv_mp3.hdf
these files contains the mp3s of the dataset and the labels.
Similar to audioset you can use:
# Example call with all the default config:
python ex_openmic.py with trainer.precision=16 -p # with 2 gpus:
DDP=2 python ex_openmic.py with trainer.precision=16 -p