search

zhaojw1998 / DAT-CVAE

Codes and MIDI demos of ISMIR 2022 paper: Domain Adversarial Training on Conditional Variational Auto-Encoder for Controllable Music Generation
MIT License
19 stars 1 forks source link

Could you please share the preprocessing of POP909 dataset? #1

Open TeresaTsang opened 1 year ago

zhaojw1998 commented 1 year ago

For the pre-processed data used in this work, you may download it here. I don't have the pre-processing code though. This file contains ~2K melody+chord pieces collected from Nottingham and POP909. Those from POP909 are in the second half. Hope this is helpful!

TeresaTsang commented 1 year ago

For the pre-processed data used in this work, you may download it here. I don't have the pre-processing code though. This file contains ~2K melody+chord pieces collected from Nottingham and POP909. Those from POP909 are in the second half. Hope this is helpful!

Thank you so much for your help! There are 2154 pairs in the .npy. I want to ensure whether the data from POP909 are in [1077:]?

zhaojw1998 commented 1 year ago

For the pre-processed data used in this work, you may download it here. I don't have the pre-processing code though. This file contains ~2K melody+chord pieces collected from Nottingham and POP909. Those from POP909 are in the second half. Hope this is helpful!

Thank you so much for your help! There are 2154 pairs in the .npy. I want to ensure whether the data from POP909 are in [1077:]?

I asked the data creator. It turns out that the second half of data.npy actually came from a superset of POP909. Hence it could not guarantee a one-to-one correspondence with pieces in POP909. If you are strictly looking for melody-chord pairs of POP909 anyway, you may refer to its official repo. Each piece has time-aligned chord annotations in chord_midi.txt. And you can use mir_eval.chord.encode() to convert chord labels to chroma vectors.