What is LUFS?

The short answer

LUFS stands for Loudness Units relative to Full Scale. It is a single number that describes how loud a piece of audio feels to a listener over time, measured on a standardized scale so that everyone, every meter, and every platform agrees on what the number means. It was defined by the broadcast standards ITU-R BS.1770 and EBU R128, and US broadcast calls the exact same unit LKFS. One unit of loudness, one LU, is worth one decibel, so the two scales line up: turning a program down by 3 LU is turning it down by 3 dB.

The key word is feels. LUFS is not the height of your tallest sample, which is what a peak meter reads. It is closer to the sensation of loudness you would report if someone asked you how loud a track was: an average over time, weighted so that the frequencies your ears are most sensitive to count for more. That is why two songs can share the exact same peak and still sit many units apart in loudness, and why LUFS, not peak, is the number streaming and podcast platforms use to match every track to a common volume.

Why peak and plain average miss it

The obvious ways to measure level both fail to match perception, each for its own reason. A peak meter reports the single tallest sample in the signal, an instantaneous height in dBFS. That is exactly the right tool for one job, making sure you never clip, but it says nothing about loudness. A track can have one lone spike touching -0.1 dBFS and be quiet the whole rest of the way, while a dense, squashed master sits near that same peak continuously and feels far louder. Same peak, very different loudness.

A plain average of the signal power, an RMS reading, gets closer, because it accounts for the whole program rather than one spike. But it treats every frequency as equally loud, and your ears do not. A deep bass rumble and a midrange voice at the same measured power do not sound equally loud: the voice sounds much louder, because human hearing is far more sensitive in the midrange than in the low bass. An honest loudness measure has to account for that, and a flat average does not.

K-weighting: counting the way you hear

The fix is a filter applied before anything is averaged, called K-weighting. It is a simple two-stage shaping of the signal that approximates the frequency response of human hearing. The first stage is a gentle high-shelf that lifts the higher frequencies by a few dB, modelling the acoustic effect of the head. The second is a high-pass filter down near 38 Hz that rolls off the deep low end. The net result is that low bass counts for less toward the loudness number, and the midrange, where speech and most instruments live, counts for close to its full value.

This is why a heavy sub-bass line can be shaking your desk while the loudness meter barely reacts: the meter is deliberately discounting the very lows, because your sense of loudness discounts them too. Only after the signal has been K-weighted does the meter measure its energy and average it. Everything a LUFS meter reports is loudness as heard, not loudness as measured with a flat ruler.

Three time windows: momentary, short-term, integrated

Loudness is not one number, because a program is not equally loud throughout. A LUFS meter reports the same K-weighted signal through three different time windows at once, and each answers a different question.

• Momentary (400 ms). The most reactive reading, averaged over just the last four-tenths of a second. It jumps around with every loud word and quiet gap, so it tells you what is loud right now.
• Short-term (3 s). A three-second average. It is smoother and slower, ignoring momentary spikes to tell you how loud the current passage or section is.
• Integrated (whole program). The average loudness of the entire piece from start to finish, and the single number people mean when they say a track is "-14 LUFS". This is what a platform reads and normalizes to.

The integrated reading uses two gates so that silence does not drag it down. An absolute gate ignores anything below -70 LUFS, which is effectively silence, and a relative gate then ignores anything more than 10 LU below the running average, so the quiet gaps between words or between songs do not count as part of the loudness. What remains is the loudness of the program while it is actually playing.

There is a reason a loud moment lifts the fast readings but hardly nudges the integrated one. Loudness averages energy, not decibels, so a brief loud passage counts for much more than a quiet moment counts against it, but spread across a whole program even a big spike is diluted by everything around it.

Why the numbers are negative

A LUFS scale almost always shows negative numbers, and that trips people up until they see the reference point. The Full Scale in the name is the ceiling: 0 LUFS lines up with the loudest level digital audio can represent, the same 0 you see at the top of a decibel meter. Real program material never averages anywhere near that ceiling, because a track whose average sat at 0 would have no room at all for its peaks and would be clipped to pieces. So loudness readings sit below zero, and the more negative the number, the quieter the program.

Where real material lands depends on what it is made for. Music mastered for streaming tends to sit around -14 LUFS. Podcasts and spoken word commonly target about -16 LUFS, though some platforms ask for -14. Broadcast under EBU R128 is quieter still at -23 LUFS, with US broadcast under ATSC A/85 at -24 LKFS. These are integrated targets for the whole program, and because 1 LU is 1 dB, the distance between any two of them is just that many decibels: streaming music at -14 is nine decibels louder, on average, than R128 broadcast at -23.

LUFS in patchd

Here is the honest part, because it matters for how you use a live mixer. patchd does not measure LUFS. Its channel meters read peak level in dBFS: a live per-channel readout of your loudest sample, from -60 up to 0, with silence shown as negative infinity, plus a VU-style meter fused into the fader. That is exactly the right tool for a real-time monitoring mixer, where the job is to catch peaks before they clip while you play, not to average loudness over a finished program.

LUFS is a program measurement, and it is made downstream, after your audio has left patchd. When you upload a finished recording, the streaming or podcast platform measures its integrated LUFS and adjusts playback to hit a common target, a process covered in loudness normalization. You do not set your loudness inside a live mixer; you deliver clean, well-staged audio and let the platform normalize it.

What patchd gives you is control over the two things that actually help that final number come out right: clean gain staging so every stage sits at a healthy level with headroom to spare, and a free limiter on any channel to hold your peaks under a ceiling. Because that limiter runs in a live path it uses zero lookahead and adds zero latency, so it catches sample peaks rather than reconstructed true peaks, but for delivering controlled, clip-free audio into whatever measures loudness next, that is exactly what you want. The difference between the peak number patchd shows you and the loudness number a platform reports is worth understanding on its own, and that is the subject of LUFS versus dBFS.

The takeaway

LUFS is loudness as your ears experience it, standardized into a number you can trust across every platform: K-weighted so bass counts for less, averaged over time rather than read as a single spike, and pinned to a full-scale ceiling so the numbers run negative, with real material sitting around -14 for streaming music, -16 for podcasts, and -23 for broadcast. The momentary, short-term, and integrated readings are the same program seen through shorter and longer windows, and the integrated one, the whole-program average, is what the world normalizes to.

patchd is a pre-launch Windows audio mixer. It will not print a LUFS number for you, because loudness is measured on the finished program downstream, but it gives you the clean gain staging and the free, zero-latency limiter that let you hand that program off with controlled peaks and room to breathe. patchd is in development now. Join the waitlist and we will tell you the moment it is ready to install.