LUFS vs dBFS

The short answer

dBFS (decibels relative to full scale) is a peak measurement. It looks at your audio one sample at a time and reports the single loudest one. It answers a yes-or-no safety question: did any sample get too close to the digital maximum, the point where it would clip?

LUFS (Loudness Units relative to Full Scale) is a loudness measurement. It listens to the whole program over time, weighted to match how your ears actually respond, and reports how loud it feels. It answers a completely different question: on average, across the whole thing, how loud is this?

Because they ask different questions, they give different answers. Two clips can both peak at exactly -0.1 dBFS, one hair below the ceiling, and yet one can sound twice as loud as the other. Peak tells you nothing about loudness, and loudness tells you nothing about peak. You need both, and you need to know which is which.

Two rulers, two questions

dBFS is instantaneous. Digital audio is a stream of samples, and each sample is a number with a hard maximum: full scale, defined as 0 dBFS. Every real sample sits below that, at some negative dBFS value, and the peak dBFS of a clip is just the highest one it ever reaches. Nothing about timing, density, or how long the sound lasts enters into it. One stray sample decides the number. That is exactly what you want from a safety meter, because it only takes one sample over the line to clip.

LUFS is the opposite kind of measurement: it is an average over time, built to track perception rather than raw amplitude. Before it measures anything it runs the signal through K-weighting, a two-stage filter (a gentle high-shelf lift plus a high-pass near 38 Hz) that makes low bass count for less than midrange, the way your own hearing does. Then it integrates the weighted energy over a window. The standards behind it, ITU-R BS.1770 and EBU R128, define three windows: a momentary 400 ms window, a short-term 3 s window, and the integrated value over the whole program. The integrated value uses two gates (an absolute gate at -70 LUFS and a relative gate 10 LU below the running average) so that silence and quiet gaps do not drag the number down.

Both scales read as negative numbers below a 0 reference, which is why they get mixed up. But the numbers mean different things. A dBFS peak of -0.1 means one sample nearly touched the ceiling. An integrated loudness of -16 LUFS means the program, over its whole length and weighted for your ears, sits well below full scale on average. One is a single tallest moment; the other is the felt loudness of everything.

Same peak, different loudness

The clearest way to see the gap is to put two clips side by side that reach the identical peak. Below, both waveforms just touch the same dashed peak rail, so a peak meter would read them the same. But one is sparse, with tall transients and quiet gaps, and the other is dense, riding near the top almost the whole way across.

This is why a whispered word and a wall of noise can share a peak. The whisper might have one sharp consonant that spikes to the ceiling and then near-silence around it, while the noise sits pinned near the top the whole time. Same peak, wildly different loudness. The area under the waveform, not its tallest point, is what your ears add up.

Why peak alone tells you nothing about loudness

Peak and loudness are not just different numbers; they move on separate axes, and a surprising amount of audio processing is about moving one without moving the other. A limiter is the clearest example. It clamps every peak down to a ceiling, so the peak axis stops at, say, -1 dBFS no matter what. But once the peaks are pinned, you can raise the overall gain to fill the space right up to that ceiling, and raising the whole signal lifts its average energy. The peak does not budge past the ceiling, but the loudness climbs. That is loudness going up while peak stays flat: two independent axes.

This decoupling is the entire mechanism behind the loudness war, and also behind why it ended. For years, masters were limited and pushed ever louder, trading dynamic range for a higher average level, because a louder-sounding track grabbed attention. Then streaming and podcast services began doing loudness normalization: they measure the integrated LUFS of your program and adjust playback gain to a common target, turning loud masters down (and sometimes quiet ones up).

Because the platform normalizes by loudness, chasing a higher average just gets turned back down on the way to the listener. The squashed master and the dynamic one arrive at the same loudness, but the squashed one gave up its punch to get there. Understanding that LUFS and dBFS are separate axes is what makes this obvious: you cannot win a loudness contest that is measured in LUFS by pushing a number measured in dBFS.

Which number to watch, and when

Watch dBFS peak for safety. It is the number that tells you whether your signal is about to clip on the way out, and it is the one a live meter should show you in real time as you set levels. If your peak is comfortably below 0, you are not clipping, full stop. This is also where true peak comes in: the reconstructed analog waveform between samples can rise a little above the tallest digital sample, so a delivery ceiling around -1 dBTP leaves room for those inter-sample peaks to stay legal through conversion and codec.

Watch integrated LUFS when you care how your program sits against a target. Those targets are published by the services and standards bodies: music streaming lands around -14 LUFS, spoken-word and podcasts around -16 LUFS (some platforms use -14), EBU R128 broadcast is -23 LUFS, and ATSC A/85 is -24 LKFS. LKFS, the unit US broadcast uses, is the same as LUFS: 1 LU equals 1 dB. If you master a podcast to a peak of -1 dBFS but leave the loudness at -24 LUFS, it is perfectly safe and perfectly quiet, and it will get turned up next to everything else. Peak was never the number you needed there.

Peak and loudness in patchd

It is worth being precise about what a live mixer actually measures, because it is only ever one of these two axes. In patchd, each channel meter shows peak level in dBFS: a live tracked dB readout that runs from about -60 up to 0 (with silence shown as negative infinity), plus a VU meter fused into the fader. That readout is the peak axis, the real number from the left side of everything above. It is there so you can set a healthy level and see instantly if a channel is about to clip.

patchd does not have a LUFS meter or a loudness meter, and that is by design, not an oversight. LUFS is a program-loudness measurement, integrated over a whole finished program with gating and K-weighting, and it is measured downstream by the streaming or podcast service that normalizes your audio, not moment to moment in a live monitoring path. A live mixer's job is the peak axis: keep every channel clean and un-clipped in real time. The loudness axis belongs to the platform. Claiming a live meter measures loudness would be claiming it does something it cannot.

The same honesty runs through patchd's limiter, a free node with a default ceiling of -1.0 dB. Because it has zero lookahead and adds zero latency, it catches sample peaks, not true peaks, which is the right tradeoff for a path you are hearing yourself through live. The engine runs about 10.7 ms through the bus at a 512 buffer, and smaller ASIO buffers take it lower, so nothing in the chain gets between you and your own voice.

patchd is a pre-launch Windows audio mixer with a free effects rack on every channel and a Studio tier at $39.99 per year; the real-time engine is identical on both. If you want a mixer that shows you an honest peak readout on every channel and never pretends to be something it is not, join the waitlist and we will tell you the moment it is ready to install.