Since this was posted in early 2016 Spotify have both implemented an album mode and lowered their loudness target by 3dB.
This is the first of a series of loudness experiments with two goals. First; we wanted to hear for our selves how different online platforms handle different loudness targets. Second; hopefully this will make it easier for anyone to understand loudness and if you produce music you’ll get a better idea of how your desired loudness target will affect your audio on current distribution platforms. This will enable you to make decisions based on knowledge rather than a perception of a loud vs. a dynamic master in a quick A/B comparison (where I would say about 90% choose the louder version whether or not it actually sounds better).
This is a WIP. Measurements and more releases will follow…
Most music retailers such as Spotify and Youtube loudness compensate their content so that you as a listener don’t have to change your volume control every time you change song or video clip, and some don’t (Soundcloud for example). This experiment gives you the opportunity to listen to the same track mastered at different loudness targets on the most common current music platforms and to the raw files that where uploaded. We recommend these steps:
- Download the raw files and extract them: https://drive.google.com/open?id=0B1rylQZfCdreU05PRXdHZFRzZG8
- Load all songs in a player that do not compensate for loudness. We recommend Foobar for Windows and VLC on OSX.
- Compare the raw files with the audio from online retailers
- Bonus: There’s level matched versions in the LUFS.rar archive so that you easy can hear how the loudness targets affected the sound.
Links to the most common platforms:
We distributed the album to lots of common online stores but choose to link to the most interesting ones when it comes to loudness. If you’re interested you easily find the album by searching for Saintpid or LUFS.
Details about the experiment
We took one track (Truck Driver by Noisebud) and mastered it to 10 different loudness levels, all with the limiter set to -1dBFS. We named the tracks according to their dynamic range (DR) and loudness values (LUFS), and released them as one album with the 10 versions as is first, then with all versions level matched to make it easier to compare audio quality. The track in itself is pretty dirty (Noisebud call their music electronic punk) so this is no audiophile experience. That’s why we plan for this to be a series of releases, with styles ranging from pop to metal, singer songwriter to techno, and other electronic genres.
We’re searching for material so if you want to contribute please contact us at info(at)saintpid.se. You get your song mastered for free but that’s the only compensation we can offer at this point.
Here’s a Key-value report from the software Toscanalyser.
The interesting values are the RMS average, RMS max, Peak max, Peak average, Sub Bass (SB%) and the K-system values (K12-20).
The chart show some notable facts.
- The amount of low-end has almost been cut in half in the loudest version vs the most dynamic one (1.1% vs 2.0%).
- The last version that show an OK K12 value is the DR5 -8.6LUFS version. We should’ve stopped here if not before.
- The last version that gets a safe K12 value is the DR8 -12LUFS version. This is the version the artist (Noisebud) prefer.
Here’s the DR report from the offline Dynamic Range Meter:
Thanks to Sigurdór Guðmundsson (http://skonrokkstudios.com/) and Ian Stewart (http://ianstewartmusic.us/) for ideas, consultation and trying to push the track to DR1 (which Sigurdór finally did but it’s not on the album).
This is the first more thorough measurement. We did Spotify first because we needed the data for a meeting this Monday where SSES (Swedish Sound Engineers Society http://www.sses.org) and Spotify having a sit down to discuss loudness normalization. Or rather, we’re going to make a plan on how to discuss the issue in public during the LLB audio fair in Stockholm next month. Hopefully, we’ll manage to influence Spotify to bring down their loudness target to the AES recommended -16LUFS (http://www.aes.org/technical/documents/AESTD1004_1_15_10.pdf).
Since this article were posted early 2016 Spotify has lowered their loudness target by 3dB and implemented an album mode. We leave the measurements here for historical reasons but strike everything that no longer apply.
We measured the tracks in both “Album mode” and in “Playlist mode”. We did that because it makes sense to loudness normalize tracks from different releases in a playlist but if you play an album you would prefer the internal gain differences between tracks to be intact as the artist, producer or mastering engineer intended. To be fair, there are situations when you might want to loudness compensate within an album. For example in a noisy environment or listening on tiny tiny speakers, but the way to go would be an extra user setting specific for those situations.
Sadly, Spotify currently adjusts the loudness independently even in album mode.
The Album as a whole (Integrated BS.1770-2/3)
With Normalization turned on:
Peaks -2.11dBTP (-2.13dBFS)
Measures from the album playback (we only measured the same three as for the playlist mode)
“DR2 -4.2LUFS” measures -10.2LUFS with peaks at -4.36dBTP
“DR5 -8.6LUFS” measures -10.8LUFS with peaks at -1.33dBTP
“DR12 -16LUFS” measures -13.2LUFS with peaks at +0.92dBTP
Spotify doesn’t seem to take playback type into account, the internal gain structure of an album will be changed by Spotify in the same way as if the tracks were placed in a playlist.
The recorded waveforms of “DR2 -4.2LUFS”, “DR5 -8.6LUFS” and “DR12 -16LUFS” in both Album and playlist mode measures the same.(scroll down for the screenshots of quick measurements for Youtube and iTunes)
We will do more thorough measurements of the following streaming services as soon we have the time. Meanwhile, Sigurdór recorded the whole playlist from the most common places and you can see the results here:
Youtube perform their level normalization within a day or so after the file being uploaded. My guess is that they leave it as is at first so when people check the upload they don’t get confused when the level dropped by half. These screen captures are taken after the level normalization kicked in. It’s the -16LUFS to the left and the -4.2LUFS to the far right. We could’ve used 8 bits for the loudest version and saved some bandwidth. =)