your loudest master never arrives loud
streaming turns your master down to a target before anyone hears it, so mastering louder buys nothing but lost dynamics. and the loudness war mostly did not do what people think: measured across five decades the spread between loud and quiet did not narrow. what came down was the transient crack, by about a decibel and a half. so the thing to watch is heavy limiting and clipping, more than compression as an idea, and the move is to master until it is finished and let the platform set the number.
streaming turns your master down to a target before anyone hears it. two masters of the same song, one squashed as hard as it goes and one left alone, both get pulled to the same loudness on the way in. the loud one arrives with everything it gave up to get there, and none of the loudness.
the number you are aiming at
the figure you have seen is minus 14 LUFS. it is real, and it is spotify’s, in their own words. but even on spotify it is one of three: quiet is minus 19, normal is minus 14, loud is minus 11, and your listener picks which. that is an 8 dB spread you do not control.
then the other four. apple evens songs out but does not publish to what. youtube has a feature called stable volume and says it is on by default. amazon normalizes. tidal, i could not find a current document at all. so “master to minus 14” is aiming at one company’s number and guessing the other four.
the war mostly did not happen
the fear is that everything got crushed flat. measured across five decades, the spread between loud and quiet moments did not narrow, and the change was not statistically significant. what came down was crest factor, how far a transient sticks up above the body of a song, by about a decibel and a half. the war flattened the hits and left the arc of a song alone.
and preference is a separate question from quality. twenty mastering engineers on their own gear picked the more compressed version 17 times out of 24, and whether they could hear it tracked how peaky the music already was, more than how much was taken off.
what to do with it
master until it sounds finished. the platform sets the playback loudness either way, and the number you would chase lives in a listener’s settings menu.
watch your transients and your clipping before you watch a LUFS meter. a survey of masters found 79% still come in above minus 14 a decade after normalization arrived, fewer than half are free of clipping, and the mixes going in were cleaner than the masters coming out. heavy limiting is the thing to be careful about, more than compression as an idea. i do not have a clean answer for how loud is loud enough. i have a cleaner one for what to stop chasing.
none of these numbers are mine, and i say so on camera. everything i measured is in the video, both standards documents linked in the description. kernaudio.io
more field notes
Sep 8, 2026
·sound science
the dialogue you mixed is not the dialogue they hear
a chain of reasonable decisions ends at the sofa with the voice buried. the film is about the chain. this issue is about your end of it: how to mix dialogue so it survives a stereo fold-down, a flat panel and an ordinary pair of ears, with the numbers i use, including one my own films failed on.
Sep 1, 2026
·sound science
a row of spikes nobody trained in
generated music leaves a mark, and it has nothing to do with taste. it is a row of evenly spaced spikes, put there by an operation that is also sitting inside half the plugins you own. which gives you a rule you can use on any analyzer: evenly spaced peaks came from a process, musically spaced peaks came from the music. plus a second tell the paper does not cover, and a note on why neither one is worth trying to EQ.
Aug 25, 2026
·sound science
nobody publishes where a phone speaker stops
two films, a few days apart, and the same thing happened twice. i went looking for the frequency a phone speaker gives up at so i could mix against it, and it isn't published anywhere. then i sat down to check the tuning arithmetic and found the vallotti preset in almost every synth is named after the wrong person. go and check a number, and it's either not published or it's filed under someone else's name.