What Is Loudness Normalization? LUFS Explained

Loudness normalization raises or lowers a whole recording so that it sounds as loud as a chosen target. The target is measured in LUFS (Loudness Units relative to Full Scale). LUFS doesn't look at the single highest peak. It measures how loud the audio sounds to a listener over time, so after normalization a quiet voiceover and a loud one play back at about the same perceived volume.

For narrated screen recordings, about -16 LUFS integrated is a good default. If you record in Screen Bolt, one switch, Volume Normalization, brings your microphone track to that level.

Loudness vs volume

Volume is a level setting, such as the gain on a track or the slider on your speakers. Loudness is what you actually hear. Two files can have the same peak level and sound very different. Picture a voice recorded far from the mic with one loud cough. It can peak at the same level as a close, steady voice, yet the steady voice sounds much louder.

A peak meter can't tell those two files apart. A loudness meter can, because it averages the signal over time and weights it the way ears do.

This matters for screen recordings because people watch them back to back. If your tutorial is much quieter than the video before it, the viewer has to reach for the volume and can miss your first sentence.

What LUFS measures

LUFS is the unit defined by the ITU-R BS.1770 standard. US broadcast documents call the same measurement LKFS, and the two are identical. A loudness meter does three things:

1. Filters the audio with a "K-weighting" curve. The curve plays down deep bass and slightly favors the upper frequencies, which roughly matches how sensitive human hearing is.
2. Averages the energy over a time window.
3. Ignores silence when it measures integrated loudness. It first drops blocks quieter than -70 LUFS. It then drops anything more than 10 LU below the average of what's left. Because of this, pauses between sentences don't drag the number down.

Values are usually negative because they're measured relative to digital full scale (0 dBFS, the loudest a digital file can go). So -14 LUFS is louder than -16 LUFS.

One loudness unit (LU) equals 1 dB. If you raise a file's gain by 3 dB, its loudness goes up by 3 LU.

Loudness meters show a few related readings:

When someone says "normalize to -16 LUFS", they mean integrated loudness.

Platform loudness targets

Broadcasters and streaming services normalize playback so listeners don't have to keep adjusting the volume. These are the common reference points as of November 2026:

Two things follow for screen recordings.

Many places your video ends up don't normalize at all. A file shared in Slack, embedded in Notion or a help center, placed on a landing page or kept in a shared folder plays at the level you exported. The loudness you deliver is the loudness people hear.

Don't count on a platform to fix a quiet file. Where normalization exists, its job is mainly to keep loud uploads from blasting. A voiceover that went out too quiet is likely to stay too quiet.

Why -16 LUFS works for voice recordings

A screen recording with narration is closer to a podcast than to music, so the spoken-word target fits. At -16 LUFS integrated, your audio:

• Matches Apple's podcast spec, so your voice sits at a level people are used to hearing for speech.
• Leaves headroom, so peaks don't need heavy limiting to stay under -1 dBTP.
• Lands within 2 LU of the -14 LUFS usually quoted for music services and YouTube. If one of them turns the file down, the change is small.

If you publish mainly to YouTube and your voice is already even and dense, -14 LUFS is also reasonable. If the file goes to several places (docs, Slack, social, YouTube), -16 is the safer single choice.

Loudness normalization vs compression

People often mix these up, but they solve different problems.

Loudness normalization applies one gain change to the entire file. Every part goes up or down by the same amount, so the gap between your quiet and loud sentences stays the same. Only the overall level moves.

Peak normalization is the older approach. It raises the file until its single highest sample hits a set level, such as -1 dBFS. One mouth click or desk bump can then decide the gain for the whole recording. That is why peak-normalized voiceovers can still sound quiet.

Compression changes the dynamics. When the signal goes above a threshold, the compressor turns that part down by a ratio. At 3:1, every 3 dB over the threshold comes out as 1 dB over. Loud words come down and the gap between quiet and loud parts shrinks. Compression doesn't aim for a target loudness.

In practice you use both. Compression evens out the voice, and loudness normalization puts the evened-out voice at the target. If you normalize an uneven voice without compression, the average is right but some sentences are still too quiet and others too loud.

Normalizing a screen recording's voice

In Screen Bolt

Screen Bolt records your microphone as its own track, separate from system audio, so your voice can be processed on its own. Volume Normalization sets the gain of the mic track to about -16 LUFS integrated. Voice Clarity adds compression.

1. Open your recording in the editor.
2. Open the Audio panel (⌘4, Command-4).
3. Under Audio Enhance, turn on Volume Normalization.
4. Optional: also turn on Voice Clarity. It applies a high-pass filter at 100 Hz, a presence boost around 3 kHz, and compression of about 3:1 from -18 dB.
5. Wait while the editor shows Enhancing audio… NN%, then play the recording back.
6. Export as usual. The enhancement is included in the exported file. To turn everything on at once, use Auto Audio Enhance. It switches on Noise Reduction, Volume Normalization and Voice Clarity together. After that the three switches work independently, so you can turn any one of them back off. All of them are off by default, and the processing runs on your Mac.

A few details are easy to miss:

• Only the microphone track is processed. System audio, imported music and sound effects aren't normalized. Set their levels with the System Audio volume slider, each audio clip's Volume, and the volume controls in the Sounds panel.
• The finished mix can measure louder than the voice alone. Each layer you add on top (system audio, music, click sounds) adds energy. Keep music well under the voice. Adding music to a screen recording covers how.
• Exported raw mic audio doesn't include enhancement. If you right-click the microphone row on the audio track and choose Export Microphone Audio…, you get the unprocessed track as .m4a, without enhancement or volume changes.
• Long recordings can show uneven progress. The first export after you turn on enhancement for a long recording may show progress in jumps. This is expected.

Audio Enhance is available in Screen Bolt's free version. Free exports carry a watermark and are limited to 3 minutes at 720p. The Free vs Pro comparison shows what the license adds.

With ffmpeg (free, for any video file)

If you already have a finished video from another recorder, ffmpeg can measure and normalize it. ffmpeg doesn't come with macOS. If you use Homebrew, install it with this command (brew uninstall ffmpeg removes it again):

brew install ffmpeg

To measure a file without changing it, run the command below. It prints a summary with integrated loudness, loudness range and true peak:

ffmpeg -i input.mp4 -vn -af ebur128=peak=true -f null -

To write a normalized copy targeting -16 LUFS with true peaks capped at -1.5 dBTP, run the command below. It leaves the original file untouched, copies the video stream without re-encoding, and re-encodes only the audio:

ffmpeg -i input.mp4 -af loudnorm=I=-16:TP=-1.5:LRA=11 -ar 48000 -c:v copy output.mp4

Keep -ar 48000 in the command. Without it, loudnorm outputs audio at a 192 kHz sample rate.

Two limits apply here. In a single pass, loudnorm adjusts the gain as it goes, so it acts more like a gentle automatic leveler than one fixed gain change. For a strict single gain change, measure the file first, then run a second pass with the measured values and linear=true. Also, this command treats the whole soundtrack (voice, music and system sounds) together. It can't process the voice on its own the way a separate mic track allows.

When normalization doesn't fix the sound

Normalization only moves the level. It can't repair the recording. If the result still sounds wrong, check these causes:

• Background hiss or room noise got louder. Raising a quiet recording to -16 LUFS raises the noise with it. Turn on noise reduction as well; see removing background noise from a screen recording. If macOS Voice Isolation was on while you recorded, Screen Bolt skips its own noise reduction for that recording because the mic was already denoised. Using Voice Isolation on Mac explains the setting.
• Some sentences are still too quiet. This is a dynamics problem, not a level problem. Add compression (Voice Clarity in Screen Bolt), or keep a steadier distance from the mic.
• The voice sounds distorted. If the original recording clipped, the distortion is part of the audio, and changing the level won't remove it. Lower the input level and record again.
• The voice sounds thin or boomy. The cause is the mic and the room, not the level. See choosing a microphone for screen recording on Mac.
• The music drowns out the voice. The voice is at the target but the music is too loud. Lower the music clip's volume instead of pushing the voice higher.

FAQ

Is LKFS different from LUFS? No. Both names describe the same ITU-R BS.1770 measurement. LKFS appears mostly in US broadcast specs, and LUFS appears in European specs and most software.

Should I normalize to -14 or -16 LUFS for YouTube? Either works. YouTube doesn't publish a target, and around -14 is the figure usually quoted. -16 gives voice a little more headroom and also suits copies of the file shared outside YouTube.

Does loudness normalization reduce audio quality? A plain gain change doesn't. Problems come from side effects: noise in a quiet recording gets raised along with the voice, and peaks can clip if nothing limits them.