Lyrics Conditioning about jukebox HOT 7 CLOSED

Yes you definately can pass you own lyrics, artist and genre! Just edit the values passed in sample.py in metas!

from jukebox.

We'll try to add a command line script to pass these args to make this easier

from jukebox.

This line removes OOV characters. It's possible that two words are joined together resulting in a word the model has not seen before. You can normalize the lyrics yourself or try sampling more.

from jukebox.

Amazing thank you waiting for the docs then. Will try sample.py in the meanwhile!

from jukebox.

@prafullasd Thank you I managed to condition the model through the new Colab. We have noticed that there are some “unpredictable” tokens or better not understandable words in the singing voice generation. Could this be due to the training vocabulary used for the GPT-2 model, that may have a unseen tokens in our input?

from jukebox.

@heewooj thank you, closing then.

from jukebox.

@prafullasd Thank you I managed to condition the model through the new Colab. We have noticed that there are some “unpredictable” tokens or better not understandable words in the singing voice generation. Could this be due to the training vocabulary used for the GPT-2 model, that may have a unseen tokens in our input?

You mind sharing how you do it, by condition on lyric?

from jukebox.