Possible Bug in Style Diffusion Inference Code

[brthor]

Here we see that the ground truth for the denoiser is the (acoustic_styles, prosodic_styles):

https://github.com/yl4579/StyleTTS2/blob/main/train_second.py#L307

But here we see that the output from the sampler is parsed as (prosodic_styles, acoustic_styles)

https://github.com/yl4579/StyleTTS2/blob/main/train_second.py#L739-L743

Also during inference the output from the sampler is parsed as (prosodic_styles, acoustic_styles):

https://github.com/yl4579/StyleTTS2/blob/main/Demo/Inference_LibriTTS.ipynb?short_path=4b85bf5#L275-L289

I tried swapping them and got bad inference results, so I'm not sure what is going on with this.

/cc @yl4579