The master needs not to be clipped, and the easiest way to check this is to look at the True Peak, which takes into account inter-sample peaks. A bit of googling shows that Spotify is specifying a max True Peak of -2dB for loud tracks, which is really conservative and, frankly, suggests that their encoder is sub-optimal.
This can be trivially done by service providers in automated ways rather than blaming mastering engineers because streaming services providers decide what formats and settings to use, not mastering engineers. Different codecs and different settings create different peaks, in some worst cases the encoded files can have 10dB over as illustrated by iZotope:
Ideally, Spotify should just normalize the masters to ensure the encoded files don't contain 0dBFS+ values by using a two-pass approach, if what they use are not floating-point compliant.
That might pose a problem preserving the intended relative loudness of songs when playing albums though, since this normalization would effectively be done per song. That’s also going to take a substantial amount of processing over the back catalog on Spotify’s end. Additionally, Spotify’s actual volume normalization results in 16-bit output and is limited to -1 dBFS, before the volume control. IMO it would be better to have the limiter after the volume control—so with normalization on, the volume control essentially sets a target loudness, while allowing the full digital dynamic range above that to be used as well for more dynamic songs that need it. All of these goals could be most easily facilitated by just having the Ogg decoder output in float.
Of course, the most ideal ways are using floating point in every step, and have more flexibility at the client side normalization settings. Did they not earn enough to implement these things? I wonder.
Probably a lack of interest if anything. The audio pipeline has probably largely stayed the same from when Spotify was a tiny startup more than 10 years ago. I can’t imagine that many people petitioning Spotify to have their Ogg decoder output float instead of 16-bit integer, so here we are. I thought about posting a suggestion in their community forums, but I have no idea how I would effectively communicate this idea to non-technical people.
If the level is reduced by 18 dB then it means that the album was mastered at -5 LUFS. Such album, at least in my opinion, does not deserve 16 bits in the first place
This is trueIf the level is reduced by 18 dB then it means that the album was mastered at -5 LUFS. Such album, at least in my opinion, does not deserve 16 bits in the first place
Albums mastered at -23 LUFS (e.g. classical) will retain all 16 of their bits. Albums mastered at -14 LUFS will lose only 1.5 bit.
Well this was while I was in the gym using bose ear buds so..
