I don't see why this needs to be done at the track mastering level though. If audio device manufacturers know their users are primarily going to be listening in noisy environments with cheap headphones they can compress the track at playtime.
Then, as an end user, we could choose which we preferred. By default there would be no change in behaviour. Where software updates are available, an the option could be provided to swap between versions at will. For older devices, the streaming service could let the end user choose the high dynamic range version as an account-level default.
Personally I’d want ready access to both versions. When listening to music is the singular activity in a quiet environment, full dynamic range is great. But as soon as I’m not solely focused on music, multitasking or in a noisy environment, I’d actually prefer the compressed version.
In the old days, the recommendation for digital audio was to master for -20 dBFS average, use no compression (or very little) and let any peaks fall where they may in the 20 dB headroom. And nobody complained about the noise floor at ~76 dB below the average level.
I wish everyone would go back to mastering to that spec, instead of slamming everything to 0 dBFS with loads of compression and often clipping on top. Obviously still allowing use of compression for artistic reasons.
The bandwidth consequence would be trivial bordering on nil given that only a small number of albums would ever receive the treatment, and only a small number of end users would choose the option.
(If it did matter, which it doesn’t, it occurs to me that you could even create a 24 bit file that only contains 18 significant bits [dithered to 18 bits then padded to 24] in such a way that the space was reclaimed by ALAC or FLAC encoding.)
[0] https://www.sweetwater.com/insync/what-is-lufs-and-why-shoul...
Many games let you specify the audio output, and then adjusts the mixing accordingly. This would be something similar I guess, but the mixing is probably best done at the streaming provider, to save bandwidth and other resources. (For static content like music it only really has to be done once per output type.)
I'd like something like this, I have a vastly different sound setup at home than on the go, but listen to mostly the same tracks. Most sound pretty good on my home setup, but on the go I often find myself adjusting volume up and down between tracks, despite things like normalization being on.
* https://en.wikipedia.org/wiki/ReplayGain
(I have no knowledge in this field, so am asking out of genuine curiosity.)
It's not possible to separately compress constituent stems in a mix. Once they're mixed, the impact that you can have on "loudness" is restricted to the whole, frequency-separated bands (multi-band compressor), or extracted parts. From an information-theory perspective, the available leeway for control over presentation is restricted once the audio is mixed.
However, I'd rather have a simple compressor running on playback than have a whole industry ruin their content.
To be fair, I believe mastering is one of those domains that may soon be made obsolete with machine learning. But we're not quite there yet, and the existing methods require a lot of DSP power.