Professional Analog Mastering.
Professional Analog Mastering.
226K+ subscribers
For most music, a 2-channel model of this standard is used. The left and right inputs are fed into independent K-weighted filters. The average loudness or RMS is then measured for each channel. A gate is then used to remove anything below -70LUFS from the measurement.
The signals are summed, multiplied by a logarithmic function to give each channel a specific value (in a stereo track, the left and right channels are valued identically), and, lastly, a second gate negates anything 10dB below the current measurement.
Most streaming services use this standard when measuring a master’s LUFS. So, if we want our tracks to sound louder, our ultimate goal here is to make the meter measure our master as quieter than it subjectively sounds.
For example, if we can make the master sound like it’s -9 LUFS, but the meter measures it as -10 LUFS, then when it’s normalized a target loudness of -14 LUFS, it’ll be turned down roughly 4dB instead of 5 dB. As a result, our master will sound 1dB louder than others.
When determining the loudness, K-weighting is used to approximate the loudness at which we perceive frequencies more closely.
To emulate how we hear low frequencies as quieter, a 12dB/octave HP filter is set around 100Hz. And to emulate how we perceive high frequencies as louder, there’s a 4dB shelf amplifying 2kHz and above by 4dB.
Figuring out how to optimize a track for this standard took a lot of testing, so before I show you the primary way to increase perceived loudness after normalization, I want to cover
Looking at this K-Weighted chart, you’d assume, or at least I did, that removing imperceivable or barely perceivable frequencies would lower the LUFS.
For example, if you have a master, and you attenuated everything below 30Hz, and everything above 20kHz, it makes sense that removing these ranges would lower the LUFS.
Even Spotify claims that some tracks sound louder or quieter after normalization because these frequencies cause less accurate LUFS readings.
But everything I measured shows that it makes no difference with the integrated LUFS and barely changes any other metrics.
So, What Makes A Track Sound Louder on Spotify?
Or any other streaming service, for that matter?
The leading cause is a higher true peak level.
For example, the track Guess by Charlie XCX and Billie Eilish sounds incredibly loud, even after normalization.
Without normalization, the LUFS is -7.5. Since dB and LU have a 1:1 ratio in the K-Weighted ITU standard, the peak level for this track normalized to -14 LUFS should be -6.5 dB. However, the normalized version is peaking around -3.5dB in most places and even up to -2.7 at one point.
So what’s happening here?
Although the standard measures true peaks, it doesn’t measure the distortion they cause during playback.
Since this distortion is dynamic or only occurs when a transient surpasses 0dBTP, the distortion increases the loudness whenever a transient hits.
As a result, the kick, snare, and any other aggressive and often percussive transients sound louder, even after normalization, since the distortion is not accounted for in the initial measurement used for normalization.
That’s a big reason this supposed -14 LUFS normalized track measures as -12.5 LUFS when recorded into the DAW.
The distortion occurs when it’s played back, and a new track measurement reveals it’s 1.5LU higher than it should be.
If you’re mastering an aggressive genre, one in which distortion is already a part of the sound, know that if you avoid true peak limiting and let the true peak level go above 0dB, this additional level will cause distortion that increases the loudness of the track.
It’s not the best choice for every genre, but in some, the added distortion doesn’t sound bad by any means.
This has more to do with tracking and mixing, but it has big implications for mastering, too.
To better understand how the current loudness standard measures different frequency ranges, I divided white noise and pink noise into five ranges each.
White noise is linear; all ranges are equal in amplitude. Pink Noise is logarithmic or emphasizes lows and de-emphasizes highs to more closely match how we perceive the loudness of each frequency range.
For a white noise track peaking at -5dB, -6.6LUFS is the measured loudness overall; notice how the lows play a smaller role in the LUFS than the highs.
Anything above 1kHz causes the meter to measure the track as louder, especially when 10kHz and above are equal in amplitude to other ranges.
The pink noise track, also peaking at -5dB, is measured at -11.3LUFS overall. Even though the highs are attenuated significantly, and the lows are amplified, the highs still play the most significant role in the measured loudness.
Anything above 1kHz contributes most to the measured loudness of the track.
Does this mean that when mixing or mastering, we should attenuate highs or boost lows more to increase perceived or subjective loudness but not measured loudness?
In short, no. But, this info, paired with a rarely discussed psychoacoustic effect, could be the key to understanding increasing subjective loudness while circumventing measured loudness.
I don’t know if this effect has a name or to what extent it’s been studied, so I’m just going to call it
This is the idea that the listener anchors the loudness of a song around a vocal - this isn’t to say the vocal’s frequencies exactly, just the recognition of a vocal itself.
If most listeners are geared toward trying to discern what the vocalist is saying or singing, then if the vocal is covered over, masked, or competing with too much instrumentation, most listeners will perceive the overall track as quieter because the primary element, the vocal, can’t easily be heard.
Combining this idea with what we learned about how high ranges create higher measured loudness and taking a closer look at Guess, the track we measured earlier, I found something interesting.
When the vocal performance is most complex, which in this case occurs in the verses, the vocal is the only melodic instrument in the high range.
This causes the overall high range to be lower in amplitude since, again, the vocal is present but not much else.
Now, to the listener, it sounds loud - the vocal can be heard and is upfront. But to the LUFS meter, it’s quiet. The primary ranges that contribute to measured loudness are sparse and lower in amplitude. This causes a lower overall LUFS, even though the listener doesn’t perceive it as quieter.
To demonstrate this, I’ll play two tracks with identically measured LUFS. For the first track, the vocal will be lowered while the instrumental is amplified.
For the second track, the vocal will be amplified while the instrumental is lowered. Again, the meter says the 2 have the same loudness, but I want to know if one sounds subjectively louder to you.
Watch the video to learn more >
Cutting extreme lows or highs doesn’t cause a loudness meter to measure the LUFS as lower, at least not in any meaningful way.
The distortion caused by True Peaks increases the perceived loudness and circumvents the measured loudness by occurring after measurement.
LUFS meters prioritize high frequencies, even more than the K-Weighting filter would suggest.
And lastly, reducing melodic info while a vocal is present causes increased subjective loudness while reducing measured loudness.
When all these techniques and concepts are combined, a song can sound significantly louder than another, even if both are normalized to the same LUFS.