Understanding the "Settings/Audio/Audio Info" Chain (ReplayGain, dither, etc.)

[EthelTheFrog]

Hi. I'm hoping someone can help me understand some aspects of the process depicted in the Audio Info section.

I can't seem to take screenshots on my DAP, so I'll just describe the relevant features of what I'm seeing:

I'm starting with a 16 bit FLAC file. Under the DSP, I see a conversion to a Float64/32 bit. Then, under "Output", I have a "Direct HD" 24 bit since that's what I've manually set under the Output settings.

My primary question is: at what stage of this process is ReplayGain applied? I'm thinking of this question in regards to any distortions it could have on the audio. For example, I know that if you take a 16-bit file and digitally de-amplify it, you risk pushing those samples with amplitudes at the bottom-most steps down to the noise floor because you'll run out of bits. I read somewhere that if you take a full-range (peaking at 0 dB) 16-bit file, digitally de-amplify it by -18 dB, and save it as 16-bit, you're effectively encoding a 13-bit file.

As far as I understand it, this is essentially what ReplayGain is doing (digital de-amplification), so, if the gain is applied directly to the 16-bit file, one would probably see some information loss. (Whether one could hear it is a different story.) I'm not sure how the floating-point nature of the converted file (under "DSP") affects things, but, with the much larger bit depth afforded by the 32 bit structure, I can't imagine running into any information loss if applied at that stage in the process.

For the "Output" stage, I originally did not have the "Hi-Res Output" option enabled and so it was outputting a 16-bit file. My thinking went: even if there was no information loss at the DSP step, if it's then down-converting back to 16-bit, it won't actually avoid the issue because it's just restricting the bit-depth again at the end. The 24-bit structure has such a larger bit-depth that, according to my napkin math, you would need an extreme value for your ReplayGain tag to yield any information loss when applied to a 16-bit source file.

So, I'm supposing that, if the ReplayGain is applied at the DSP stage, then there shouldn't be any information loss with this setup.

On a related note, is there dither applied between the "DSP" and the "Output" stages (32-bit -> 24-bit)? I don't see any mention of it and it seems like a lot of processing for the machine to be doing on the fly, but it would seem to me an odd omission especially when it was originally outputting a 16-bit file to the DAC.

Please let me know if I've misunderstood something here. I'm relatively new to these audio engineering concepts. Thanks.

[andrewilley]

ReplayGain is applied as part of the DSP process (so within the internal 64/32 float stage). The point of it is usually to reduce the gain rather than increase it, but in the case of a very low level recording it could go the other way. 

Andre