Literally just a Kazoo

Description

I'll reupload this to Huggingface when I figure out how to use it but for the time being have a Mega link. The dataset doesn't contain any copyrighted material, I just played a plastic kazoo and got different people to play it in front of my mic to have samples of different pitches/styles, recorded it then trained the AI on that. I recommend using mangio-crepe and setting the index to 1 otherwise the model will start pronouncing words like its a voice. *NOTE: For the samples below I used acapellas of Californication and Golden Hour but I recommend recording the reference audio yourself by humming or singing "duh duh" as you would with a kazoo. This is just so the end result ends up sounding realistic.*