turbo@mvsep.com
Piano validation dataset consists of 50 tracks, each around one minute long and the sample rate is 44100. Piano stem includes acoustic piano as well as electric piano. Only mixed tracks are available for download. Stems for piano and other are closed and are used to check the quality of the separation. Dataset size ~0.9 GB.
Download Piano dataset (~0.9 GB)
To do this, you need to download the dataset. Run your algorithm on this dataset. Each of the 50 melodies should be divided into 2 parts: piano and other. This way you should get exactly 100 files. The naming is as follows:
As an example for file: song_piano_016_mixture.wav you should get 2 files:
1) song_piano_016_piano.wav - must contain the piano part of the composition
2) song_piano_016_other.wav - must contain all other stems from the composition
You can upload the result in .wav, .flac and .mp3 formats.
Guitar validation dataset consists of 30 tracks, each around one minute long and the sample rate is 44100. Guitar includes acoustic guitar as well as electric guitar. The dataset has a closed part, which consists of audio files containing only the instrumental part of the composition and containing the guitar part of the composition. This part is hidden for correct check of the algorithms on the side of our server. Dataset size ~0.5 GB.
Download Guitar dataset (~0.5 GB)
To do this, you need to download the dataset. Run your algorithm on this dataset. Each of the 30 melodies should be divided into 2 parts: guitar and other. This way you should get exactly 60 files. The naming is as follows:
As an example for file: song_guitar_016_mixture.wav you should get 2 files:
1) song_guitar_016_guitar.wav - must contain the guitar part of the composition
2) song_guitar_016_other.wav - must contain all other stems from the composition
You can upload the result in .wav, .flac and .mp3 formats.
Medley Vox validation dataset consists of 380 mono mixtures with 44100 sample rate. Each mixture contains multiple singers perform at the same time. You need to separate each mixture in 2 parts: vocals1 and vocals2. The rules:
1) The most loud vocal must go to vocals1. Most loud is measured as mean of absoulte value for the whole waveform.
2) If there are more than 2 vocals in the mixture you need to put the most loud vocal in vocals1, while all others must be put in vocals2.
Dataset size ~0.4 GB.
Download Medley Vox dataset (~0.4 GB)
To do this, you need to download the dataset. Run your algorithm on this dataset. Each of the 380 tracks should be divided into 2 parts: vocals1 and vocals2. This way you should get exactly 760 files. The naming is as follows:
As an example for file: medley_016_mixture.wav you should get 2 files:
1) medley_016_vocals1.wav - must contain the most loud vocal for the mixture
2) medley_016_vocals2.wav - must contain all other vocals from the mixture
You can upload the result in .wav, .flac and .mp3 formats.
Strings validation dataset consists of 30 tracks, each around one minute long and the sample rate is 44100. Strings stem includes only bowed strings. Only mixed tracks are available for download. Stems for strings and other are closed and are used to check the quality of the separation. Dataset size ~0.9 GB.
Download Strings dataset (~0.6 GB)
To do this, you need to download the dataset. Run your algorithm on this dataset. Each of the 30 melodies should be divided into 2 parts: strings and other. This way you should get exactly 60 files. The naming is as follows:
As an example for file: song_strings_016_mixture.wav you should get 2 files:
1) song_strings_016_strings.wav - must contain the strings part of the composition
2) song_strings_016_other.wav - must contain all other stems from the composition
You can upload the result in .wav, .flac and .mp3 formats.
Wind validation dataset consists of 35 tracks, each around one minute long and the sample rate is 44100. Wind stem includes brass as well as woodwind instruments. Only mixed tracks are available for download. Stems for wind and other are closed and are used to check the quality of the separation. Dataset size ~0.9 GB.
Download Wind dataset (~0.6 GB)
To do this, you need to download the dataset. Run your algorithm on this dataset. Each of the 35 melodies should be divided into 2 parts: wind and other. This way you should get exactly 70 files. The naming is as follows:
As an example for file: song_wind_016_mixture.wav you should get 2 files:
1) song_wind_016_wind.wav - must contain the wind part of the composition
2) song_wind_016_other.wav - must contain all other stems from the composition
You can upload the result in .wav, .flac and .mp3 formats.
DnR v3 Test dataset consists of 104 tracks, each one minute long and the sample rate is 48000 in FLAC format. It's subset of DnR v3 dataset test data, which very large in original form. We choose several samples for every language presented in original dataset. Dataset size ~0.5 GB.
Download DnR v3 dataset (~0.5 GB)
To do this, you need to download the dataset. Run your algorithm on this dataset. Each of the 104 melodies should be divided into 3 parts: music, sfx and speech. This way you should get exactly 312 files. The naming is as follows:
As an example for file: song_dnr_016_mixture.flac you should get 3 files:
1) song_dnr_016_music.wav - must contain the music part of the track
2) song_dnr_016_sfx.wav - must contain the sfx from the track
3) song_dnr_016_speech.wav - must contain the speech from the track
You can upload the result in .wav, .flac and .mp3 formats.
Super Resolution Checker for Music (SRCM) consists of 50 tracks, each one minute long and the sample rate is 44100 in WAV format. It contains music melodies which high frequences were cut at (1000, 2000, 3000, 4000, 5000, 6012, 7000, 8000, 9000, 11000, 12000, 14000, 16000, 18000), some tracks contains only vocals. Your goal is to restore original track with all high frequnces enabled. Dataset size ~0.9 GB.
Download Super Resolution Checker for Music dataset (~0.9 GB)
To do this, you need to download the dataset. Run your algorithm on this dataset. Each of the 50 melodies should be restored. This way you should get exactly 50 files. The naming is as follows:
As an example for file: song_sr_016_mixture.wav you should get 1 file:
1) song_sr_016_restored.wav - must contain the restored track
You can upload the result in .wav, .flac and .mp3 formats.
turbo@mvsep.com