Any plans for ReplayGain analysis to use the recent EBU-R128 standard ? [DONE]

[Steven]

all default settings, and the value adjusted so its compatible with the old ReplayGain
In the following MB version i plan to remove support for the old replay gain algorithm

[Bee-liever]

Thank you Steven
That puts a lot of my concerns about the new analysis and also Mr Belkner's use of the ReplayGain tags to rest.

[Zak]

And I don't think they use the same reference volume level.

http://wiki.hydrogenaudio.org/index.php?title=ReplayGain_2.0_specification#Reference_Level_and_Gain

You're right. I thought RG1 used 89 too, but it's definitely 83.

all default settings, and the value adjusted so its compatible with the old ReplayGain

Does that mean tracks analysed with RG2 will be based on a reference of 83dB to match the volume from tracks already analysed with RG1?

In other words: will there be a noticeable difference in volume between tracks analysed with RG1 and RG2 / should tracks analysed with RG1 be re-analysed with RG2 to keep the volume of the whole library consistent?

[Bee-liever]

The new replaygain being used is not ReplayGain2.0 [RG2] but EBU-R128. The RG2 specification is still a 'work-in-progress' and not finalised.
The lib1770.dll being used is a distribution of R128GAIN.  An EBU-R128 compliant loudness scanner.

And I don't think they use the same reference volume level.

http://wiki.hydrogenaudio.org/index.php?title=ReplayGain_2.0_specification#Reference_Level_and_Gain

You're right. I thought RG1 used 89 too, but it's definitely 83.

It's to do with different measurements, but RG1 does equate to an 89 dB reference level.

all default settings, and the value adjusted so its compatible with the old ReplayGain

Does that mean tracks analysed with RG2 will be based on a reference of 83dB to match the volume from tracks already analysed with RG1?

I'm assuming here, but I hope that means that R128GAIN only has the '--rg-compatible' option set.  That means they get analyzed to the new EBU-R128 standard and then 5dB (I think) is either added or subtracted (can't remember which) to align it with RG1 settings.

In other words: will there be a noticeable difference in volume between tracks analysed with RG1 and RG2 / should tracks analysed with RG1 be re-analysed with RG2 to keep the volume of the whole library consistent?

From what I have read, the volume should be around the same, but the perceived levels of bass and treble should be improved.
Pat will have a better understanding on this, as I have gleaned much of this info from his posts on hydrogenaudio.

[redwing]

That means they get analyzed to the new EBU-R128 standard and then 5dB (I think) is either added or subtracted (can't remember which) to align it with RG1 settings.

I think that's what Steven meant by "the value adjusted" part.

From what I have read, the volume should be around the same, but the perceived levels of bass and treble should be improved.

With my library,
- Only 0.2% of the tracks remain the same in Track Gain value after re-analysis.
- But the degree of the changes were smaller than 3.0 dB for 98.2% of the library. The change was smaller than 2 dB for 71% of the library, and smaller than 1 dB for 30.9% of the library.
- 83.2% of tracks gained an average of 1.69 dB while 16.5% lost an average of 0.72 dB. Overall, the average of track gain increased to 0.01 dB from the previous -1.28 dB with the difference of 1.29 dB level.

[hiccup]

From what I have read, the volume should be around the same, but the perceived levels of bass and treble should be improved.
Pat will have a better understanding on this, as I have gleaned much of this info from his posts on hydrogenaudio.

Some other important strong-points of EBU-R128 (which I think have not yet been pointed out here), are that the algorithm is also tuned to address specific challenges in spoken word and movies/commercials (including surround sound channels).

So anyone playing mixed content such as audio-books, music, TV-series etc. on his system will experience far more extreme advantages using EBU-R128 vs. the old RG1.

[Anti]

Can I ask what you guys set the 'adjusted calculated replaygain values' setting to when doing volume analysis with this new algorithm? From what I've read in this thread, 0dB is now more like a 83dB reference level (as opposed to the previous 89dB reference)? So to avoid clipping, should it still be set to -1 or -2, or even lower?

If it makes any difference to the answer, I listen to 'progressive metal' which can vary multiple times in a single track from very, very soft and quiet to the most extremely loud and compressed music possible. I stopped using the old replaygain, because even setting the 'adjustment' parameter to -3 or -4dB, I would still get an alert saying that replaygain adjustment would cause clipping on some tracks. The alert doesn't tell you which tracks caused the alert, or how much clipping it would cause, so I simply stopped using it. I was hoping this new algorithm would fair better?

[hiccup]

Setting the MusicBee slider to -5dB would approach the EBU-R128 recommendation of -23 LUFS.
That will still not prevent clipping in all cases. You have to remember that the resulting proximate 84dB is just an average perceived loudness value, so that could still result in amplifying peaks into clipping.
Even the (currently) lowest MusicBee slider value of -6 will not be sufficient (by far) for many cases.

Some rule of thumb: if you have plenty amplifier or headphone power (sensitivity) -5dB would probably be advisable.
On a high-end system with some power-reserve probably even lower, so the volume difference with tracks that have been attenuated even more to prevent clipping will be smaller.

On a sidenote, the new algorithm optionally provides for a 'true-clipping' sampling mode.
Extremely critical audiophiles would want to use that mode to be 100% sure any clipping is completely ruled out.
The drawback is that analyzing is substantially slower though.
(edit: don't go searching in MusicBee for this option, it's not implemented)

[junebaug]

may I hook on the smart-gain-question of Pingaware and the answer of Simon? As I understand the smart gain feature:

first rule: it takes album gain if in the file, otherwise falls back to track gain if in the file.
second rule: if play mode is shuffle: smart gain directly goes for track gain (but if AutoDJ is active it acts like in first rule)

Well: first rule is simple

Problems arise with second rule: IMHO album gain is only a good choice if I listen to whole albums - or at least if 2 tracks of the same album follow each other in a playlist. But with properly mixed playlists - or in AutoDJ-mode - I don't see the point why smart gain goes for album gain here? I would expect going for track gain.

Could some one please shed a bit light on this?

Regards
 Ralf

[Steven]

if the auto-dj is behaving that way then its an oversight that i will correct. I will check that in the next few days

[junebaug]

Thanx Steven for having a look into this.

What do you think: could this be an oversight too, that, if I put together a playlist manually - without AutoDJ - than it is the same behaviour: the mixed tracks are played with album gain and not with track gain.

Regards
 Ralf

[Steven]

Smart gain will now use Track gain values when the Auto DJ is enabled and now when shuffle is not enabled MB also checks that at least the following track is from the same album before using Album gain values. When shuffle is enabled and grouped by Album then Album gain values will be used, otherwise Track gain values

[junebaug]

Sounds pretty good to me 

Thanx Steven!!

[Steven]

http://musicbee.niblseed.com/V2_4/MusicBee_Exe_Patched.zip

[Pat]

Awesome.