Training what text ?
steinhaug opened this issue · comments
OK, this is definately interesting if its possible to expand further on existing models. But....
Say I prepare 100 audio clips, wav 35 sec each, would it not be practical for all that I also have some text ?
File: the wav file
Genre: musical genre, eg. electro house
Creativedesc: a description describing the music, ex: a bad ass beat that comes falling from the roof... Upon impact opn the floor a bassline explotion appears while a steady electro bass beat is rolling.
Desc: A badass bassline, electro house beat starting a song.
I am thinking it would be nice to add (atleast be able to) a creative description, which is where I find this AI intersting. And surely a more "straight up music" approach. Note that Google also have a text2music AI and there aswell they recommend long creative descriptions. (after all how else are we supposed to have different tracks).
Maby a json format ? Typically a .json file with same name as audio file.