Anyone with the horrible distorted audio for wav/mp3...

[DarkAlchy]

The issue is the dev wrote a bad maths routine. The vae already gives the proper output so this code:

                output = (
                output.to(torch.float32)
                .div(torch.max(torch.abs(output)))
                .clamp(-1, 1)
                .mul(32767)
                .to(torch.int16)
                .cpu()
            )

Made it min, max, 0 in the output wav, and if you change that, in infer.py, to this:

            output = (
                output.to(torch.float32)
                .div(torch.max(torch.abs(output)))
                .clamp(-1, 1)
                .cpu()
            )

it should cure it as it did for me.