What does mastering actually change in a track?

Mastering · 2 min read

On 12 premaster/master pairs, mastering did three things every time: it made the track about 11 LU louder, it pulled the peaks about 6 dB closer to the loudness, and it left the loudness range almost where it was. The tonal balance moved in a different direction on almost every track.

We measured each song's mix and its released master; the method and every number are at the end.

Every master is louder than its mix, by 6.4 to 13.3 LU. True peak rose by only 3.1 to 6.3 dB. Dot: the median of the 12; bar: lowest to highest.

Louder, every time

The mixes sat 6 LU apart; the masters land in a band 3.3 LU wide, around -10 LUFS. One LU is one dB of loudness, as ITU-R BS.1770 defines it, so 11 LU is a fader pushed up 11 dB. What makes it mastering is the peaks.

The peaks did not follow

Pure gain would raise true peak by the same 11 dB. It rose by 4.4. So the peak-to-loudness ratio (PLR) fell in all 12 songs: transients were brought down against the body of the sound, which is what limiters and compressors do.

Every master stops short of full scale, the loudest at -0.83 dBTP; the club records on Club loudness, measured behave differently.

Loudness rose 11 LU and PLR fell 6 dB, while LRA barely moved. Bar: the median change; line: smallest to largest of the 12.

The song kept its shape

Loudness range (LRA), as EBU Tech 3342 defines it, compares quiet and loud passages in 3-second windows: verse against chorus, breakdown against drop. It barely moved. The transients got squeezed; the arrangement did not.

The tone moved, but not in one direction

Change in each band's share of the energy. Bar: the median change; line: smallest to largest of the 12. Every band went up in some songs and down in others.

In 5 songs, some band moved by more than 15 points. The only weak pattern: the high band rose in 9 of 12, from a tiny share. The plain reading is that each mix asked for something different; no fixed "mastering curve" would reproduce these 12.

What this does not tell you

n is 12: hip hop, pop, house and rock. Too few per genre, and no techno or minimal.
One pack, unknown engineers. Possibly one studio's style: closer to one opinion than twelve.
Not a recipe, not a quality score. "+11 LU" describes these songs, not a target for yours, nor which version sounds better.

How we measured, with every number

Method. 12 pairs, 24 WAV files (PCM, 24-bit, 44.1 kHz, stereo): for each song, the mix before mastering and the master that was released. Source: a free practice pack published by an online mastering-tools vendor, 3 songs each of hip hop, pop, house and rock. Engineers behind the masters: unknown. Measured with our own analyzer, which measures loudness with libebur128 through pyebur128 0.1.1 and the spectrum with librosa 0.11.0. Integrated loudness and true peak per ITU-R BS.1770; loudness range per EBU Tech 3342. Comparison: each master minus its own premaster, then the median and the range of those 12 differences.

The changes. Every row is a difference, master minus premaster. "Up in" counts how many of the 12 songs moved upward.

Metric	Unit	Median change	Smallest	Largest	Up in
Integrated loudness	LU	+10.97	+6.43	+13.28	12 of 12
True peak	dB	+4.42	+3.10	+6.31	12 of 12
Peak-to-loudness ratio (PLR)	dB	-6.23	-10.09	-1.43	0 of 12
Loudness range (LRA)	LU	-0.71	-2.10	+0.25	4 of 12
Sub share, 20-60 Hz	pp	-5.15	-12.20	+28.90	5 of 12
Bass share, 60-250 Hz	pp	+1.50	-9.50	+29.70	6 of 12
Mid share, 250-4000 Hz	pp	+3.00	-34.80	+19.50	8 of 12
High share, 4-20 kHz	pp	+1.15	-10.70	+2.60	9 of 12
Spectral centroid	Hz	+63	-1027	+1118	6 of 12

Where the two sides start and land.

Metric	Unit	Premaster median	Premaster range	Master median	Master range
Integrated loudness	LUFS	-21.03	-23.11 to -17.12	-10.09	-11.47 to -8.18
True peak	dBTP	-5.42	-7.34 to -4.24	-1.03	-1.23 to -0.83
PLR	dB	15.59	11.13 to 18.87	9.06	7.26 to 10.44
LRA	LU	6.04	2.18 to 12.54	5.53	2.40 to 11.82

Peaks. If mastering were only gain, true peak would rise by the same amount as loudness; the remaining 6 dB or so is what the peaks gave up. PLR fell by a median 6.2 dB, from a premaster median of 15.6 dB to a master median of 9.1 dB. The short, loud events (drum hits, transients) were brought down relative to the body of the sound. No song kept its premaster PLR. The loudest true peak among the 12 masters is -0.83 dBTP: a choice made on these songs, not a property of mastering in general.

Two dynamics numbers, two time scales. PLR compares the highest instantaneous peak to the loudness of the whole file. LRA, per EBU Tech 3342, is the spread between quiet and loud passages measured in 3-second windows. On these 12 songs, mastering took a lot off the first and almost nothing off the second; four of the 12 masters came out with a slightly wider LRA than their mix.

Tone. Band shares are fractions of total signal energy in four bands, without frequency weighting. Sub went up in 5 songs and down in 7. Mid went from -34.8 to +19.5 points. The spectral centroid, the frequency where the spectrum's energy balances, moved by up to 1 kHz either way, and the median moved 63 Hz. In full-range music the energy is dominated by the low end, so the 4-20 kHz band holds between 0.2% and 14.2% of the energy before mastering: a rise of one or two points is a large relative change there, and still a small part of the whole. We do not break any of this down by genre: three songs per genre is too few.

Aggregates only. Every table and figure here shows a median, a count or a smallest-to-largest range of the 12, never one song's values: the pack is public, so a per-song row would point at an identifiable released song.

More limits. The master may come from a later mix than the premaster we have; that and mastering EQ look the same here. Energy shares are not loudness per band: they are unweighted energy.