ByteDance

/

Make-An-Audio-2

like 7

I very much am grateful that this model is a lot more simple than AudioLDM 2, but it still might be a good idea to evaluate it too, especially given the comparison to AudioLDM (1). I know after so many hours were put into this work that it is unfortunate to not be SOTA, but I still feel it is proper. This architecture seems a lot more resource efficient and data-efficient.