• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Spotify and loudness normalization - how can normalizaed track have higher DR than non-normalized

danadam

Addicted to Fun and Learning
Joined
Jan 20, 2017
Messages
956
Likes
1,496
Vorbis is a pretty decent codecs
First, by "differences" I only meant my non-perfect null.
Second, the fact that vorbis is clipping does not at all imply that it is not decent. All lossy codecs induce clipping on dynamically compressed material. Here is "Love and Peace or Else" from "How To Dismantle An Atomic Bomb" by U2 (DR4), from top to bottom: original flac, vorbis, aac (fdk), opus (lossy codecs all at 320 kbps):
lossy.clipping.png

the DR score difference is too big to be explained by lossy compression artifacts.
I stand by my theory that the TT meter is wonky.
Maybe.

Just one thing, earlier you wrote "Because the loudness range is identical". I'd be careful when drawing any conclusions from that. AFAIK loudness range involves quite more complex calculations and won't be swayed by non-linear change (e.g. clipping) in a few samples. DR value on the other hand is a rather simple formula and very sensitive to the max peak value (actually, 2nd max). For example, I can take the above track and reduce its volume by 20 dB and both loudness range and DR value will remain the same. Then I can modify 2 samples and this will not affect loudness range but will change DR from 4 to 10.

Oh, btw, I am talking here about the original DR algorithm, not the new (and proprietary) MAAT DROffline.
 
Last edited:

dasdoing

Major Contributor
Joined
May 20, 2020
Messages
4,208
Likes
2,673
Location
Salvador-Bahia-Brasil
the limiter is always on,
so when a not normalized track hit's the limiter (-1dB true peak) it is "compressed".
the same track wont hit the limiter when normalized cause it is most likely above -14LUFS integrated and will be atenuated
 

Soundstage

Active Member
Joined
Jan 10, 2020
Messages
294
Likes
216
the limiter is always on,
so when a not normalized track hit's the limiter (-1dB true peak) it is "compressed".
the same track wont hit the limiter when normalized cause it is most likely above -14LUFS integrated and will be atenuated
Assumption or finding?
Is this limiter limited to Spotify or also on Tidal?
 

dasdoing

Major Contributor
Joined
May 20, 2020
Messages
4,208
Likes
2,673
Location
Salvador-Bahia-Brasil
Assumption or finding?
Is this limiter limited to Spotify or also on Tidal?

well, that's the only explanation. they also recomend keeping everything below -1dB true peak https://artists.spotify.com/faq/mastering-and-loudness#what-audio-file-formats-does-spotify-support.
I have seen the same happening in a youtube video before.
with that beeing said, I doubt the dynamic diference is audible. modern limiters will only cut dynamics "on paper". it will be manly transients
 

AnalogSteph

Major Contributor
Joined
Nov 6, 2018
Messages
3,334
Likes
3,278
Location
.de
Second, the fact that vorbis is clipping does not at all imply that it is not decent. All lossy codecs induce clipping on dynamically compressed material.
I'd rather say they increase peak levels, at which point it is up to the decoder and playback chain to provide sufficient processing headroom for the output to not clip. Foobar2000 and Rockbox with some negative gain (using ReplayGain will generally provide more than enough) will fit the bill with MP3s, for example.

I agree with those saying something in the analyzer is either not counting >0 dBFS peaks or working on clipping decoded audio, which effectively amounts to the same.
 

Sukie

Addicted to Fun and Learning
Forum Donor
Joined
Jul 29, 2020
Messages
928
Likes
1,467
Location
UK
I am correct in thinking that Spotify only allows for Track normalisation rather than Album normalisation? It's a deal-breaker me as Track normalisation leads to ridiculous volume jumps on a quieter instrumental track. Album normalisation smooths this out.
 

danadam

Addicted to Fun and Learning
Joined
Jan 20, 2017
Messages
956
Likes
1,496
I am correct in thinking that Spotify only allows for Track normalisation rather than Album normalisation? It's a deal-breaker me as Track normalisation leads to ridiculous volume jumps on a quieter instrumental track. Album normalisation smooths this out.
In album view it uses album normalization, in playlists it uses track normalization.

And yes, I also think that track normalization is a poor idea. Fortunately for me, I don't use spotify that often and when I do it is always in album view.
 
Last edited:

Wombat

Master Contributor
Joined
Nov 5, 2017
Messages
6,722
Likes
6,459
Location
Australia
Is it possible that expansion is used in the normalisation process to compensate for volume reduction? Similarly, compression for volume increase?

Maintaining apparent audibility?
 

danadam

Addicted to Fun and Learning
Joined
Jan 20, 2017
Messages
956
Likes
1,496
Is it possible that expansion is used in the normalisation process to compensate for volume reduction?
No, just a negative gain is applied.
Similarly, compression for volume increase?
Yes, but more precisely it is a limiter set to engage at -1 dBFS.

From https://artists.spotify.com/faq/mas...-is-loudness-normalization-and-why-is-it-used
  • Negative gain is applied to louder masters so the loudness level is at ca - 14 dB LUFS. This process only decreases the volume in comparison to the master; no additional distortion occurs.
  • Positive gain is applied to softer masters so that the loudness level is at ca - 14 dB LUFS. A limiter is also applied, set to engage at -1 dB (sample values), with a 5 ms attack time and a 100 ms decay time. This will prevent any distortion or clipping from soft but dynamic tracks\
 

Wombat

Master Contributor
Joined
Nov 5, 2017
Messages
6,722
Likes
6,459
Location
Australia
It may not be just negative gain. Psychoacoustics.

In the absence of accepted standards, anything goes it seems.

Where is the transparency on what the service provider manipulations do?
 

danadam

Addicted to Fun and Learning
Joined
Jan 20, 2017
Messages
956
Likes
1,496
It may not be just negative gain. Psychoacoustics.

In the absence of accepted standards, anything goes it seems.

Where is the transparency on what the service provider manipulations do?
Not sure what you mean. Anyway, digital capture of Phil Collins / Face Value (2016 Remaster) / In the Air Tonight - 2015 Remaster ( open.spotify.com/track/6MjfEIHOMW6MaDO3LpFcmW?si=0A9omG9jSLG9wyKa7acrwA ), both with normalization disabled and enabled (in album mode):
Code:
Peaks:
 -0.24 dBFS - 1rec.none.wav
 -4.10 dBFS - 1rec.albm.normal.wav

  Loudness,     LRA,   True peak
-15.9 LUFS, 14.2 LU,    0.979083, 1rec.none.wav
-19.8 LUFS, 14.2 LU,    0.627752, 1rec.albm.normal.wav
(0.979083 is -0.18 dBFS, 0.627752 is -4.04 dBFS)

DR         Peak             RMS       Duration  Title [codec]
----------------------------------------------------------------------------------------------
 DR10    -0.24 dB        -16.21 dB      5:29    1rec.none.flac
 DR10    -4.10 dB        -20.08 dB      5:29    1rec.albm.normal.flac
After applying -3.8606 dB gain to the version without normalization, it nulls perfectly with the version with normalization:
Code:
             Overall     Left      Right
Pk lev dB     -90.38    -90.38    -90.38
RMS lev dB    -98.77    -98.77    -98.77
RMS Pk dB     -98.27    -98.27    -98.29
 

Wombat

Master Contributor
Joined
Nov 5, 2017
Messages
6,722
Likes
6,459
Location
Australia
Not sure what you mean. Anyway, digital capture of Phil Collins / Face Value (2016 Remaster) / In the Air Tonight - 2015 Remaster ( open.spotify.com/track/6MjfEIHOMW6MaDO3LpFcmW?si=0A9omG9jSLG9wyKa7acrwA ), both with normalization disabled and enabled (in album mode):
Code:
Peaks:
-0.24 dBFS - 1rec.none.wav
-4.10 dBFS - 1rec.albm.normal.wav

  Loudness,     LRA,   True peak
-15.9 LUFS, 14.2 LU,    0.979083, 1rec.none.wav
-19.8 LUFS, 14.2 LU,    0.627752, 1rec.albm.normal.wav
(0.979083 is -0.18 dBFS, 0.627752 is -4.04 dBFS)

DR         Peak             RMS       Duration  Title [codec]
----------------------------------------------------------------------------------------------
DR10    -0.24 dB        -16.21 dB      5:29    1rec.none.flac
DR10    -4.10 dB        -20.08 dB      5:29    1rec.albm.normal.flac
After applying -3.8606 dB gain to the version without normalization, it nulls perfectly with the version with normalization:
Code:
             Overall     Left      Right
Pk lev dB     -90.38    -90.38    -90.38
RMS lev dB    -98.77    -98.77    -98.77
RMS Pk dB     -98.27    -98.27    -98.29

Are you confirming what Spotify specifies what they are doing to content?

Given that small comparative sound levels are important in listening tests it seems to me when streamed content is normalised, up or down, this would be relevant and pluses and minuses re perception could be compensated for with psychoacoustic manipulation. Does it happen? Who knows?
 
Last edited:

danadam

Addicted to Fun and Learning
Joined
Jan 20, 2017
Messages
956
Likes
1,496
Are you confirming what Spotify specifies what they are doing to content?
Yes, or at least with a few tracks that I checked it was exactly like they say. And to the extent that I'm able to. I don't have access to the masters they are using so all I can compare is Spotify's client output with normalizaton on and off.

It's using album normalization for playlists too:
Tidal maybe. Not Spotify.
Here's the same "In the Air Tonight", normalization disabled compared to normalization in a playlist:
Code:
Peaks:
-0.24 dBFS - 1rec.none.wav
-0.98 dBFS - 1rec.plst.normal.wav

  Loudness,     LRA,   True peak
-15.9 LUFS, 14.2 LU,    0.979083, 1rec.none.wav
-14.6 LUFS, 13.5 LU,    0.919898, 1rec.plst.normal.wav

(0.979083 is -0.18 dBFS, 0.919898 is -0.73 dBFS)

DR         Peak            RMS        Duration  Title [codec]
----------------------------------------------------------------------------------------------
DR10    -0.24 dB        -16.21 dB      5:29    1rec.none.flac
DR9     -0.98 dB        -14.91 dB      5:29    1rec.plst.normal.flac
After reducing volume of playlist version by 1.9307 dB, it nulls with the version without normalization like this:
Code:
             Overall     Left      Right
Pk lev dB     -11.74    -11.91    -11.74
RMS lev dB    -37.37    -37.45    -37.29
RMS Pk dB     -19.29    -19.29    -19.32
So not too well. A look at the null's spectrogram and we see that it nulls perfectly in the first half of the track and half of the second half :)
gen.1rec.none.1.0__1rec.plst.normal.0.8006902786.null.png
A look at the waveforms of the null (db scale) and the track (linear scale) shows that it does not null only where the track is loudest and presumably where the limiting was applied:
in_the_air_tonight.none_vs_plst.png

So in this case, if you are bothered by the limiting but still want to have normalization, you should change the normalization level to "Quiet".
 

Wombat

Master Contributor
Joined
Nov 5, 2017
Messages
6,722
Likes
6,459
Location
Australia
So Spotify provide Specs. of their normalisation process?

Can you provide a reference?
 
Top Bottom