I think the other issue here is r8brain. the primary prupose of r8brain is to resample audio, and while it does apply dither after resampling, I don't think you can use it to only apply dither without resampling. in your case, the original files are 24/44.1, so you aren't doing any resampling, you only need to bring it down to 16 bit to get to 16/44.1.