By Keunwoo Choi
This split setting is used in my ismir paper (Automatic Tagging using Deep Convolutional Neural Networks) and my icassp 2017 submission (Convolutional Recurrent Neural Networks for Music Classification).
As stated in the paper, I used top 50 tags except few too-obviously-not-related-to-audio-content ones. Also, I only count tag as valid if the ground truth > 50. (The value ranges in [0, 100]). + Of course only those with audio files were selected. (Very recently I found out there are some zero-sized files though.)
FYI, similar models from the models in those two papers are released in https://github.com/keunwoochoi/music-auto_tagging-keras. Please take a look on it.
There are three formats - 7d, MSD, and filename. One line, one value, i.e., separated by \n. No header.
Filenames are in a format of 3/6/36122424.npy for example. Probably you want to change it as .clip.mp3 if you and I are connected by some colleagues :)
If you want to use the tags I selected, you can simply use the {train, valid, test}_y.npy as y labels.
TAGS = ['rock', 'pop', 'alternative', 'indie', 'electronic',
'female vocalists', 'dance', '00s', 'alternative rock', 'jazz',
'beautiful', 'metal', 'chillout', 'male vocalists',
'classic rock', 'soul', 'indie rock', 'Mellow', 'electronica',
'80s', 'folk', '90s', 'chill', 'instrumental', 'punk',
'oldies', 'blues', 'hard rock', 'ambient', 'acoustic',
'experimental', 'female vocalist', 'guitar', 'Hip-Hop',
'70s', 'party', 'country', 'easy listening',
'sexy', 'catchy', 'funk', 'electro', 'heavy metal',
'Progressive rock', '60s', 'rnb', 'indie pop',
'sad', 'House', 'happy']