[flac-dev] R128gain & metaflac

[lvqcl]

Ian Nartowicz wrote:

> I use it :)  To optionally allow adjustment of tracks containing different
> types of gain tag to a similar overall loudness.  Admittedly this isn't
> common.  Files using the original replaygain algorithm with a reference
> loudness of 83dB are now very rare, but R128 tags are starting to show up
> (eg. in Opus) and these play at a significantly different (quieter) level to
> tracks adjusted with the 89dB reference loudness tags.

a) According to David Robinson, "ReplayGain tags should always be quoted
relative to 89dB". So all files that were scanned 10 years ago with a
reference level == 83dB should be fixed. And IMHO it's a good idea.

b) According to http://wiki.xiph.org/OggOpus#Comment_Header
there should be no REPLAYGAIN_*** tags in Opus files; Opus uses
R128_TRACK_GAIN tag. If some audio player reads Opus tags then it should
be aware of the difference between ReplayGain and R128. But this doesn't
require REPLAYGAIN_REFERENCE_LOUDNESS tag.



> While adding the option for R128 gain into metaflac (and/or flac) would
> certainly be useful, I consider that scheme to be poorly suited to music
> playback.  I certainly wouldn't want it to be imposed, or even the default.  As
> the original, and still far more widespread, replaygain quickly discovered
> using a gain specification designed for movies leaves far too much headroom and
> hence is too quiet for music, far too quiet for standard commercial rock and pop
> recordings.  R128 is even slightly worse in this respect, being more than 6dB
> (nominally 9dB, but possibly closer to 6dB in practice) quieter than
> replaygain.  What that means is around 8 of the 16 bits in a typical music
> recording are unused, and I suspect most people would (rightly or wrongly) be
> horrified if you told them all their music was suddenly 8 bits. Turning up the
> volume cannot correct for that, although applying a preamp gain could.

Are you sure? 8 unused bits require gain equal to -48 dB. Gain values for
very loud over-compressed tracks are only -10...-15dB, so only 2...2.5
bits are 'unused'.




> Music players that have adopted R128 or are planning it have often adjusted it
> by several dB to avoid the "too quiet" issue.  With no standard in place
> concerning this and R128 itself being revised it isn't at all clear just what
> metaflac should do.  Personally I think it would be better for players to
> automatically apply a preamp gain on top of R128 replay gain tags, but unless
> and until that is widely adopted, it would be unwise for encoders to reply on
> it.

AFAICS the idea was to use R128 algorithm to calculate gain, but still
write ReplayGain tags (not some new R128 tags), with the reference level
that matches the reference level of the current RG algorithm.

For example, foobar2000 uses the following formula: "RG_gain = -18 - loudness_by_R128"
in accordance to the section 3.2.2.3 in http://www.dolby.com/uploadedFiles/Assets/US/Doc/Professional/AES128-Loudness-Normalization-Portable-Media-Players.pdf
And it seems that dBpoweramp uses it too ( http://forum.dbpoweramp.com/showthread.php?34044-ReplayGain-Utility-Codec-R4 ) -- "EBU R128 used by default", "EBU R128 defaults to -18 LUFS".