LUFS vs RMS vs Peak: What Each Measurement Means

Audio delivery specs mix three kinds of numbers. ACX asks for an RMS level between -23 and -18 dB with peaks no higher than -3 dB. Apple Podcasts asks for -16 LUFS. Broadcasters ask for -23 or -24 LUFS. Because all of them are in decibels, it is tempting to treat them as one scale. They are not: each measures something different, and the same file gives three quite different readings. This article explains what each measurement does and why each standard picked the one it did.

The shared reference: full scale

All three are measured relative to digital full scale, the largest value a digital file can hold. Full scale is 0 dBFS, so practical levels are negative numbers, and -20 means 20 dB below the maximum. What differs is what is being compared with full scale: a single instant, an average of power, or an average of perceived loudness.

Peak: the highest point

A peak meter reports the highest level the signal reaches. It says nothing about how loud the audio sounds: a single click can produce a high peak in an otherwise quiet file. Peak limits exist to prevent clipping and distortion.

Sample peak

Sample peak is the largest absolute sample value in the file, in dBFS. It is what basic meters show, and it is what the ACX Check plugin for Audacity measures.

True peak

True peak estimates the highest point of the continuous waveform that a converter or decoder rebuilds from the samples. ITU-R BS.1770 measures it by oversampling, for example four times from 48 kHz to 192 kHz, and reports it in dBTP. True peak is never lower than sample peak and is often a little higher. True peak vs sample peak explains the gap, and the clipping detector finds overs between samples in your own files.

RMS: average power

RMS stands for root mean square: square every sample, average the squares, take the square root. The result follows the energy of the signal over the measured stretch, so it relates to loudness much better than peak does. But plain RMS has three blind spots:

No frequency weighting. Deep bass, which the ear hears as quieter at the same level, counts as much as the midrange where speech intelligibility lives.
No gating. Pauses pull the number down. A tightly edited chapter reads higher than the same narration with long pauses.
Different calibration conventions. In the plain convention a full-scale sine wave reads about -3 dB RMS; meters that follow the AES17 convention show it as 0 dB. That is one reason two tools can disagree by about 3 dB on the same file.

ACX specifies its level window in this plain RMS: unweighted and over the whole file. For stereo files the ACX Check plugin averages the power of both channels together.

LUFS: loudness as people hear it

LUFS (loudness units relative to full scale, called LKFS in ITU and US documents) is defined in ITU-R BS.1770, currently in revision 5. It starts from mean-square power, like RMS, then adds two things that make it follow perceived loudness much more closely: frequency weighting and gating.

K-weighting

Before measuring, the meter passes the audio through a filter called K-weighting. It has two stages: a high shelf that adds about 4 dB of weight to the upper midrange and treble, with its midpoint near 1.7 kHz, and a gentle high-pass filter that discounts deep bass. The effect is that a bright, present voice counts for more than rumble at the same electrical level, which matches what listeners hear.

Gating

The meter cuts the audio into 400 ms blocks that overlap by 75 percent, so a new block starts every 100 ms. It then discards blocks in two steps:

An absolute gate drops every block quieter than -70 LUFS, which removes digital silence and near-silence.
A relative gate works out the loudness of the remaining blocks and drops any block more than 10 LU below it, which removes most pauses and very quiet passages.

Integrated loudness is the average energy of the blocks that survive. Because pauses are mostly gated out, a narration with long silences and the same narration edited tightly measure close to the same LUFS value, while their RMS values can differ noticeably.

Channels

Each channel is measured and the energies are added, with left, right and centre weighted equally and surround channels weighted about 1.5 dB higher. A mono file is a single channel, so it reads 3.01 LU lower than the same audio copied to both channels of a stereo file. That is where the -16 LUFS stereo and -19 LUFS mono podcast convention comes from, as explained in what loudness should a podcast be.

Momentary, short-term and integrated

Loudness meters show three time scales: momentary loudness over 400 ms, short-term loudness over 3 seconds, and integrated loudness over the whole programme. Delivery targets are mostly about integrated loudness. A related figure, loudness range (LRA), describes how much the short-term loudness varies across a programme.

Side by side

Measurement	What it reports	Pauses and silence	Frequency weighting	Where you meet it
Sample peak (dBFS)	Largest sample value	Irrelevant	None	ACX Check's peak test
True peak (dBTP)	Highest point of the reconstructed waveform	Irrelevant	None	Apple (-1 dBFS true peak), EBU R 128 (-1 dBTP), ATSC A/85 (below -2 dBTP)
RMS (dB)	Average power	Counted, lowering the reading	None	ACX window of -23 to -18 dB
Integrated loudness (LUFS)	Perceived loudness	Mostly gated out	K-weighting	Apple -16, EBU R 128 -23, ATSC A/85 -24

Why ACX uses RMS

ACX states its level requirement as an RMS range and checks it with RMS-based tools: its Audio Lab analyzer reports RMS and peak levels, and the ACX Check plugin measures plain, unweighted RMS over the whole selection. ACX has never published a LUFS figure for audiobooks. There is no exact conversion between the two, because RMS counts pauses and ignores frequency balance while LUFS does neither. Two chapters at the same LUFS can sit at different RMS levels if one has more pauses than the other.

The practical consequence: when mastering for ACX, measure RMS the way ACX does rather than converting from a LUFS figure. The 5 dB window gives you room, and aiming near the middle, around -20 dB, absorbs differences between meters. You can test a chapter with the free ACX check, and the full rule set is in ACX audio requirements explained.

Why podcasts and broadcast use LUFS

Podcasts, streaming and broadcast are about matching loudness between different programmes, so they need a measure that follows perception. BS.1770 loudness is the international standard for that. Apple specifies -16 LKFS measured per BS.1770-5, EBU R 128 specifies -23 LUFS, ATSC A/85 specifies -24 LKFS, and Spotify's music normalization measures loudness according to the ITU 1770 standard. Using the same measurement from editing to playback means a file mastered to -16 LUFS in your editor should measure close to -16 in a platform's system too.

Which numbers to watch

Podcasts and online video: integrated loudness and true peak.
ACX audiobooks: RMS, peak (keep the true peak at or below -3 dBTP to satisfy either reading) and noise floor.
Broadcast: integrated loudness, or a dialogue-gated measure for US long-form content, plus true peak and often loudness range. See EBU R 128 vs ATSC A/85.

If you want to see all of these side by side for your own files, LufsReady's mastering reports list integrated loudness, loudness range, true peak, RMS, sample peak and noise floor before and after processing, so you can tell which measurement a preset is aiming at. For a quick reading without processing anything, the free LUFS meter shows integrated loudness, true peak and loudness range in your browser.

Sources

Check your file in seconds

Measure loudness, true peak and noise floor for free in your browser, or master a file to the exact target with a report.

Free LUFS meter Start free