Thank you stingray. Lots of good information. I don’t think this has been a real problem for anyone, because there are probably thousands of major label MP3 albums that were (and are still) made from 44.1 16 bit CD masters, where noise shaping had been applied in the making of the 16 bit master. Many of those MP3 albums have thousands of clip levels because they’re still made at CD level, and even that hasn’t been a real problem. That’s just the way it’s been for 10 years.
But I’m still wondering if all end user processing (including the DAC SRC, which will affect files, CDs, everything) is considered problematic post-dither processing if the DAC is upsampling from a dithered master that was made with colored dither or noise shaping. The consequence I read of not using flat TPDF is this:
“If a colored dither is used instead at these intermediate processing stages, then frequency content may “bleed” into other frequency ranges that are more noticeable, which could become distractingly audible.”
They don’t specifically mention noise shaping in that sentence, only colored dither. It probably hasn’t been a real-world problem for anyone, but if flat TPDF is a safer option all around, for any bit-reduce stage (even the one you think is the final one), and an engineer doesn’t have a preference, I think it would be helpful to know.
I’ve thought the same thing when I’ve encoded directly from 32 bit float to MP3, but I never got around to blind tests. Maybe there’s something there, even vs. 24 bit dither, I’m not sure, but I seemed to think so at the time. If you were previously dithering to 16 bit (I don’t think you said if you were previously dithering to 24 or 16) that would probably have made a bigger difference.