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

[Music1969]

Hi all

I saw this interesting page:

https://forums.sonicacademy.com/t/spotify-loudness-and-dynamic-range-ongoing-research/35300

In particular this:

Here's the official word from Spotify:

https://artists.spotify.com/faq/mas...-is-loudness-normalization-and-why-is-it-used

So, the limiter is only applied when normalization enabled - is that correct?

If that is correct, how can the normalized Dua Lipa track (Physical) have higher DR than the track with normilization turned off ?

Unless it was his DAW that was 'limiting' with the non-normalized track (peak 0dBFS), causing distortion and reduced DR? He doesn't say that but is that the only explanation here?

 

[abdo123]

Maybe the TT meter gets wonky when the ceiling is not 0 dB anymore. Because the loudness range is identical.

 

[Music1969]

Maybe the TT meter gets wonky when the ceiling is not 0 dB anymore. Because the loudness range is identical.

The loudness range of the Dupa Lipa isn't identical though - 4.7 vs 4.8 , no?

Only very slightly higher but strangely higher with the normalized version

 

[sigbergaudio]

Slightly related question: Is this normalization feature generally considered to not degrade sound quality? Based on the description it seems to be plausible?

 

[Music1969]

Slightly related question: Is this normalization feature generally considered to not degrade sound quality? Based on the description it seems to be plausible?

That's the main point of this thread.

If it actually allows greater dynamic range (compared with normalisation turned off) like my opening post suggests, then that's a positive.

 

[sigbergaudio]

That's the main point of this thread.

If it actually allows greater dynamic range (compared with normalisation turned off) like my opening post suggests, then that's a positive.

Not sure how that happens, but yes indeed.

 

[Music1969]

Not sure how that happens, but yes indeed.

All streaming services should enforce it everywhere. Give no option to disable.

It will force music makers to use dynamic range to stand out from the crowd - that's the end result we all want LOL.

At the moment you can still disable/enable in Tidal and Spotify and Apple Music etc.

 

[abdo123]

All streaming services should enforce it everywhere. Give no option to disable.

It will force music makers to use dynamic range to stand out from the crowd - that's the end result we all want LOL.

At the moment you can still disable/enable in Tidal and Spotify and Apple Music etc.

Youtube has forced -14 LUFS loudness normalization too.

 

[abdo123]

Btw, if you’re using spotify connect you’re not getting your loudness normalized.

The normalization occurs at the client, not the server.

 

[danadam]

Youtube has forced -12 LUFS loudness normalization too.

Googling "youtube lufs" says it is -14 LUFS.

 

[abdo123]

Googling "youtube lufs" says it is -14 LUFS.

It’s funny how from all the platforms, Youtube is actually becoming the new Audiophile platform.

 

[Yasuo]

I wish he would have tested a song with very high DR like 12 or 14. Only New Order - Blue Monday has an average DR. On songs with poor DR lowering the volume results in a DR increase, but we get the opposite for songs with good/great DR - or am I wrong?

 

[abdo123]

I wish he would have tested a song with very high DR like 12 or 14. Only New Order - Blue Monday has an average DR. On songs with poor DR lowering the volume results in a DR increase, but we get the opposite for songs with good/great DR - or am I wrong?

You can test whatever song you like, you don’t even need any extra gear.

 

[danadam]

I captured a digital output of Spotify's linux client for the "Physical" track ( open.spotify.com/track/4U3u4qHEAmm4Jb2J3GAfuV?si=HywbkvfKSWC6JLxLfEGzng ), both with normalization disabled and enabled. Using SoX, loudness-scanner and dr14_t.meter I got those numbers:

Peaks:
-0.00 dBFS - 1rec.none.wav
-6.89 dBFS - 1rec.albm.normal.wav

  Loudness,     LRA,   True peak
-6.5 LUFS,  5.1 LU,    1.052240, 1rec.none.wav
-14.0 LUFS,  5.1 LU,    0.455980, 1rec.albm.normal.wav

DR         Peak            RMS        Duration  Title [codec]
----------------------------------------------------------------------------------------------
DR4     -0.00 dB        -5.95 dB       3:12    1rec.none.flac
DR5     -6.89 dB        -13.42 dB      3:12    1rec.albm.normal.flac

(1.052240 is +0.44 dBFS, 0.45598 is -6.82 dBFS)

Then I reduced the volume of the track without normalization by 7.4606 dB. The result seems to null with the track with normalization, although it is far from ideal:

             Overall     Left      Right
Pk lev dB     -30.83    -30.83    -31.20
RMS lev dB    -71.52    -71.17    -71.90
RMS Pk dB     -55.45    -55.45    -56.72

That was surprising to me. A look at the null's spectrogram shows that it nulls perfectly in the first 10-15 seconds and then differences show up:

A look at the null's waveform in dB scale shows that the diffferences are at single samples once in a while. After applying 7.4606 dB gain to the track with normalization we get clipping in exactly those samples:

If not for the clipping, the new peak would be at +0.57 dBFS.

And the conclusion? Well, I don't know. Maybe that Spotify applies some kind of limiting even when normalization is disabled? That would make peaks bigger relative to RMS in version with normalization enabled and thus increase DR value.

 

[pozz]

https://www.researchgate.net/public...ations_for_Music_Streaming_Delivery_Platforms

 

[Soundstage]

View attachment 114009
https://www.researchgate.net/public...ations_for_Music_Streaming_Delivery_Platforms

Hyper compression is not only made to sound louder as the article suggests, but also to sound well in noisy surroundings.

 

[danadam]

And the conclusion? Well, I don't know. Maybe that Spotify applies some kind of limiting even when normalization is disabled?

Here's an idea, maybe those differences could be explained as a result of lossy compression producing samples above 0 dBFS. A track with so much dynamic compression as in this example will likely produce many such samples when lossy compressed. If the client decodes to float and applies volume normalizaton in float, then without normalization those above 0 dBFS samples are clipped and with normalization they become below 0 dBFS samples and are not clipped.

 

[abdo123]

Here's an idea, maybe those differences could be explained as a result of lossy compression producing samples above 0 dBFS. A track with so much dynamic compression as in this example will likely produce many such samples when lossy compressed. If the client decodes to float and applies volume normalizaton in float, then without normalization those above 0 dBFS samples are clipped and with normalization they become below 0 dBFS samples and are not clipped.

Vorbis is a pretty decent codecs, and 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.

 

[Soniclife]

Slightly related question: Is this normalization feature generally considered to not degrade sound quality? Based on the description it seems to be plausible?

Yes, it's just a change of volume in the digital domain. Depending on what's being sent the data, and how will it handles digital full scale it has the possibility of increasing fidelity, as not everything handles things correctly. I leave this turned on in Roon too, I see it as a good thing.

 

[Music1969]

Maybe that Spotify applies some kind of limiting even when normalization is disabled?

That's exactly what i hypothesised in my opening post.

The official word on Spotify using limiting is, it is on with normalisation enabled

I couldn't find any official word on limiting with normalisation off

Thanks for doing this analysis