Not an issue - question on actually conditioning generated audio

ciaua / unagan

Code for Unconditional Audio Generation with GAN and Cycle Regularization

MIT License

75 stars 17 forks source link

Not an issue - question on actually conditioning generated audio #7

Closed johndpope closed 3 years ago

johndpope commented 3 years ago

are there any efforts to actually force the audio to fit a midi sequence that you are aware?

ciaua commented 3 years ago

I think the task you described is closer to singing voice synthesis where the score (midi in your case) is given. One possible way to achieve what you described is to extract the pitches from the singing voices with tools like Crepe, and then feed the extracted pitches to the model as conditions

johndpope commented 3 years ago

Nice idea. I’ll check out crepe - thanks