Free Loudness Meter with True Peak and LRA

What LUFS measures

LUFS stands for Loudness Units relative to Full Scale. It is a measure of how loud audio sounds to a listener, averaged over time, and it is the unit streaming services use to set every track to a similar playback level. A song at −8 LUFS sounds much louder than one at −16 LUFS.

The method comes from the broadcast standard ITU-R BS.1770. The meter first filters the audio with a curve called K-weighting, which cuts the deepest bass and lifts frequencies above about 2 kHz by up to 4 dB, to follow how the ear judges loudness. Sub-bass therefore counts for less than the same energy in the vocal range.

Readings are negative because 0 is the top of the digital scale, and real music always sits below it. One loudness unit (LU) is the same size as one decibel, so a track at −10 LUFS is 4 dB louder than one at −14 LUFS.

Integrated loudness is gated. The meter splits the file into 400-millisecond blocks, ignores any block quieter than −70 LUFS, then ignores blocks more than 10 LU below the average of what is left. Silence, room tone and a long fade-out therefore do not drag the result down. This is the gating the EBU R128 broadcast recommendation uses too.

LKFS, a name used in the United States and in some specifications, is the same measurement under another name. −16 LKFS and −16 LUFS are the same level.

LUFS compared with peak and RMS meters

A peak meter shows the highest instantaneous level of the signal. It tells you whether a file will clip, and very little about how loud it sounds, because a single drum hit can reach the top of the scale in a quiet song.

An RMS meter averages the signal power over a short window, which comes closer to perceived loudness. It treats every frequency the same and has no gate, so a bass-heavy mix reads louder on RMS than it sounds, and a long quiet intro pulls a long-term RMS reading down.

LUFS adds the hearing curve and the gate to that average. Two tracks at the same integrated LUFS sound close in volume to most listeners, which two tracks at the same peak or RMS level often do not.

The six readings and what each tells you

Integrated loudness is the figure services normalize to and the one people mean when they ask how many LUFS a song is. The short-term and momentary maximums show how loud the hottest section gets, such as a chorus or a drop, which can be several LU above the integrated value.

Loudness range is measured from the 3-second readings. Following EBU Tech 3342, it ignores the quietest passages, then takes the spread between the 10th and the 95th percentile of what remains, so one very loud hit or a long fade does not inflate it. A heavily limited pop master shows a small LRA. An orchestral recording or a film mix shows a much larger one.

True peak and sample peak

Sample peak is the highest value stored in the file, in dBFS (decibels relative to full scale). A digital file cannot store a sample above 0 dBFS.

True peak estimates the waveform a converter will rebuild between those samples. The meter oversamples the audio four times to find it. The rebuilt wave can rise above the highest sample, so a master whose samples stop at −0.1 dBFS can have a true peak above 0 dBTP. Those inter-sample peaks can distort on playback and after conversion to MP3 or AAC.

Streaming loudness targets by platform

Streaming services measure each track and turn its playback volume down, or sometimes up, to a common reference level. These are normalization levels, the volume each service plays at. A track above or below the level is still released, and the service changes its playback volume to match. The services also change these levels over time, so treat the table as a snapshot.

Spotify’s levels come from its own artist documentation. The other figures are reference levels reported by audio engineers who measure the services’ players, and any of them can change with an app update.

The meter compares your integrated loudness with each level and shows the difference in decibels. The table shows what that looks like for two loud masters, which every service in the list turns down.

A master quieter than a service’s level is a different case. The meter shows how far below the level it sits, but whether the service closes that gap depends on the service and on the track’s peaks, as the next section explains.

How normalization changes playback

Spotify works in two modes. When a listener plays a whole album, it sets the gain for the album as a whole, so quiet songs stay quieter than loud ones as the artist intended. In shuffle or in a playlist, it adjusts each track on its own. Tidal uses album normalization, with the loudest track of an album setting the level.

Turning a quiet track up is harder than turning a loud one down, because raising the level can push its peaks over full scale. Spotify raises quiet masters only as far as their peaks allow, leaving 1 dB of headroom, and applies a limiter only in its Loud setting. YouTube and Tidal do not turn quiet tracks up at all.

Normalization is a listener setting. Spotify’s web player and third-party devices such as smart speakers and TVs play without it, and Apple Music listeners can switch Sound Check off. A release is heard at its original level by some listeners and at the normalized level by others.

How loud to master a song

There is no single correct loudness for a master. −14 LUFS is where Spotify and YouTube play by default, and it is a reasonable reference, but a song mastered louder does not lose anything except the dynamics the limiter took away to get there.

A loud master turned down to −14 LUFS plays at the same level as a quieter, more dynamic master of the same song. The dynamic version often sounds bigger, because its drums and transients still have room to hit. A master far below the reference, on the other hand, plays quieter than everything around it on services that do not turn tracks up.

Genre conventions vary widely. Dense electronic, hip-hop and rock masters often sit well above −14 LUFS, while jazz, folk and classical releases often sit below it. Compare your track with commercial releases in the same style, match their levels by ear, and keep the version that sounds best once the volumes are equal.

For peaks, Spotify recommends keeping true peak below −1 dBTP, and below −2 dBTP for masters louder than −14 LUFS, to leave room for the encoding to its lossy formats. The guide to how to master a song walks through the chain from EQ to the final limiter, and what a limiter does explains the ceiling setting that controls true peak.

Loudness standards for podcasts and broadcast

Speech and broadcast have their own targets. Broadcasters treat theirs as delivery requirements and can reject a program that misses them, while the Apple Podcasts figure is a recommendation.

Podcasts are often heard on phones and in noisy places, and their target sits well above the broadcast standards. Speech also needs a small loudness range, so quiet words stay audible over traffic. A light compressor narrows the range, and audio compression explained covers the settings.

Fixing what the readings show

• True peak above −1 dBTP. Lower the ceiling of the final limiter to −1 dBTP, or −2 dBTP for a loud master, and turn on its true peak mode if it has one.
• Integrated loudness far below the reference. Add a limiter or a compressor before you raise the gain, so the level goes up without the peaks clipping. Making audio louder in Audacity shows normalize, compress and limit step by step.
• Integrated loudness far above the reference. Every service will turn it down. If the master sounds flat or distorted, back off the limiter and accept a lower number.
• Large loudness range on speech. Compress or level the voice so the quiet passages come up.
• Short-term maximum many LU above the integrated value. One section is much louder than the rest. That can be the arrangement you want, or a chorus that needs taming.

Different meters can show slightly different numbers for the same file. Older versions of the standard, a different gating setting, or measuring an MP3 instead of the WAV it came from all shift the result by a fraction of a unit. Check the final upload file, and judge small differences by ear.