LUFS vs RMS

RMS (root mean square) and LUFS both describe the overall level of a signal rather than just its peaks, but they're built very differently -- and on the same piece of audio, they won't generally agree.

RMS: a straightforward average

RMS is a purely mathematical measurement: square every sample, average the squares over a window, take the square root. It has no idea what frequency it's looking at and no concept of silence -- a sustained sub-bass tone and an equally "loud" (by RMS) vocal line can sound completely different in volume to a listener, because RMS treats every frequency the same way the ear doesn't.

LUFS: modeling perceived loudness

LUFS starts from the same idea -- measure level over time, not just peaks -- but adds two corrections aimed specifically at matching human hearing:

• K-weighting shapes the frequency response before measurement, so low and high content contribute the way they're actually perceived rather than in raw mathematical terms.
• Gating excludes quiet and silent passages from the measurement, so perceived loudness reflects the parts of a track a listener would actually call "the loud parts."

Neither correction exists in RMS. That's the entire difference: LUFS is RMS-style averaging with a perceptual model bolted on, standardized (ITU-R BS.1770 / EBU R128) so every implementation measures it the same way.

Why the numbers don't match

Because K-weighting reshapes the frequency balance and gating removes quiet sections from the average, a LUFS reading and an RMS reading of the same file are routinely a few dB apart, and the gap isn't constant -- it depends on the track's frequency content and how much quiet material it has. There's no fixed conversion factor between them; you have to actually measure each one.

Why this matters in practice

RMS is still useful as a quick, cheap level check inside a DAW, but every loudness target you'll actually be held to -- Spotify, Apple Music, YouTube, broadcast delivery specs -- is published in LUFS, not RMS, precisely because LUFS was built to track what listeners perceive rather than what a simple average reports. Mixing or mastering to an RMS target and assuming it'll land at a particular LUFS value is a guess, not a measurement.

See also: What is LUFS? for the full standard, and the LUFS & dB converter to work with real numbers.