Contribution: checkpoints AVAILABLE!

[Frederico S. Oliveira]

Hi guys,
First I would like to thank @junjun3518 for the excellent work of developing and sharing the code. I trained the model following the paper settings for two weeks on a V100 GPU using ratio=2 and 3. I would like to contribute to the project by sharing the checkpoints. Below are the download links.

nuwave x2:
https://drive.google.com/file/d/1pegayKs-i78yWlPuLIp-BCU8KxxCpBzd/view?usp=sharing

nuwave x3:
https://drive.google.com/file/d/12RUMjEALAs0EoEw6Fqf9ZkpTm3COX6sf/view?usp=sharing

The following are images of the training logs.

nuwave x2:

epoch:

loss:

val loss:

nuwave x3:

epoch:

loss:

val loss:

b-a01c-8df0b99c9e0e.svg)

[Frederico S. Oliveira]

I also ran the test scripts:

nuwave_x2:

nuwave_x3:

[이준혁(Junhyeok Lee)]

Thank you for your great contribution! I add link of this issue on README.md!

[Nathan Raw]

Hey there @junjun3518 and and @freds0. Can I (or @junjun3518, if they'd like) share this model on the Hugging Face Hub?

[Edwin]

Thank you for your great contribution! I want to use checkpoint to upsample music. What performance devices do I need? Can a personal laptop run this model?

[이준혁(Junhyeok Lee)]

Yes, it is also runnable with CPU (but very slow).
For the music case, I don't recommend you to use this project since it is only trained with clean speech without music

[Edwin]

Yes, it is also runnable with CPU (but very slow). For the music case, I don't recommend you to use this project since it is only trained with clean speech without music

Thanks for getting back to me so quickly. If I use music data to train the model, and then use the model to upsample, is that feasible?

[이준혁(Junhyeok Lee)]

I think that it is different from the instrumental of the source.
For example, it is hard to apply to electronic music, since high-frequency sounds of electronic do not correlate with low-frequency sounds.
For classic instrumental or piano, it is applicable.

[Saurabh]

Would you say 700 epochs are enough for training?