What Is LUFS? Loudness Normalization Explained
LUFS (Loudness Units relative to Full Scale) is the standard unit of perceived loudness, defined by ITU-R BS.1770 and used by EBU R 128. Unlike a peak meter, it averages a whole programme: the audio is K-weighted to approximate human hearing, measured in overlapping 400-millisecond blocks, and gated so silence doesn't drag the number down. Loudness normalization is what platforms do with that number: they measure a file's integrated LUFS and apply one gain offset at playback so everything plays at a similar level, about -14 LUFS on YouTube and Spotify, -16 with Apple's Sound Check and -23 for European broadcast. On a normalizing platform, exporting louder than the target doesn't make you louder; it only costs dynamic range.
How LUFS Is Measured
ITU-R BS.1770 defines the measurement in three steps:
- K-weighting. Two filters approximate hearing: a high shelf of about +4 dB above roughly 2 kHz, and a high-pass that rolls off the low end, where energy adds less to perceived loudness than its power suggests. Channels are then summed with fixed weights: left, right and centre at 1.0, surrounds at 1.41, LFE excluded.
- Blocks. The weighted signal is cut into 400 ms blocks that overlap by 75%, so a new block starts every 100 ms.
- Gating. Blocks below -70 LUFS are discarded, then blocks more than 10 LU below the average of what remains. Long pauses and room tone drop out, so a talking-head video with gaps measures as loud as it sounds.
Integrated, Short-Term and Momentary Loudness
| Measurement | Window | Gated | Used for |
|---|---|---|---|
| Momentary (M) | 400 ms | No | Watching a meter while you mix |
| Short-term (S) | 3 seconds | No | Metering phrases and sections; the input to loudness range |
| Integrated (I) | The whole programme | Yes | Platform normalization and every delivery spec |
When a spec says “-14 LUFS” without qualification, it means integrated. A video at -14 LUFS integrated can still reach -6 LUFS short-term in its loudest moment; that is normal dynamics, not a fault.
LU vs dB, and LUFS vs LKFS
A loudness unit (LU) is a relative step: 1 LU equals 1 dB of level change, so turning a mix up 3 dB raises its loudness by 3 LU. LUFS is the absolute scale, referenced to digital full scale, which is why the numbers are negative. LKFS, the term in ATSC A/85, is the same measurement under another name. dBFS is different again: it reads single samples against full scale and says little about perceived loudness, so two files peak-normalized to -1 dBFS can sound very different in loudness.
True Peak (dBTP)
A sample-peak meter reads the largest value stored in the file. A true-peak meter reads the highest point of the waveform those samples reconstruct into, which often falls between samples and can run 1 to 3 dB higher. BS.1770 estimates it by oversampling at least four times. That is why ceilings sit at -1 dBTP (EBU R 128, Spotify, Apple Podcasts) or -2 dBTP (Netflix, ATSC A/85) rather than 0: lossy encoders add overshoot of their own, and the clipping happens in the listener’s decoder, not on your timeline.
Loudness Range (LRA)
LRA, defined in EBU Tech 3342, measures how much loudness moves rather than where it sits: the spread, in LU, between the 10th and 95th percentiles of the short-term values, after a gate 20 LU below the average. The percentiles stop a fade-out or a single gunshot from dominating. A heavily compressed promo has a small LRA; a film with quiet dialogue and loud action has a large one. For speech heard on phone speakers, a modest LRA keeps the quiet words audible.
Normalization vs Compression and Limiting
| Process | What it changes | Effect on dynamics |
|---|---|---|
| Loudness normalization | One gain offset for the whole file, chosen from its integrated loudness | None: loud and quiet parts move together |
| Peak normalization | One gain offset chosen from the loudest sample | None, but it ignores how loud the file sounds |
| Compression | Gain that changes over time, turning loud passages down | Reduces dynamic range |
| Limiting | Very fast compression that stops peaks at a ceiling | Shaves transients; a true-peak limiter enforces a dBTP ceiling |
Platforms normalize; they don’t compress your mix. That is why exporting louder buys nothing on a normalizing platform: the extra level is turned back down, and only the limiting you used to get there stays in the file. Normalization also can’t level a video internally. If music buries the voice, that is a balance problem, and the usual fix is sidechain ducking.
BS.1770, EBU R 128 and ATSC A/85
ITU-R BS.1770 is the measurement. Standards and platforms then choose targets on that scale:
| Standard | Where it applies | Target | True-peak max |
|---|---|---|---|
| ITU-R BS.1770 | Worldwide: the measurement itself | Defines LUFS/LKFS, gating and true peak | — |
| EBU R 128 | European broadcast | -23 LUFS, +/-0.5 LU (+/-1 LU for live) | -1 dBTP |
| ATSC A/85 | US television, enforceable under the CALM Act | -24 LKFS, anchored on dialogue | -2 dBTP |
Streaming platforms use the same measurement with their own targets, typically -14 to -16 LUFS for music and online video.
Platform Targets at a Glance
- YouTube: about -14 LUFS, measured rather than published; loud uploads are turned down.
- Spotify: -14 LUFS, published; true peak below -1 dBTP, or -2 dBTP for masters louder than -14.
- Apple Music with Sound Check: playback at -16 dB.
- Apple Podcasts: -16 LKFS +/-1 dB, true peak no higher than -1 dB.
- TikTok and Instagram: no published figure; -14 LUFS with -1 dBTP is a safe convention.
- Netflix: -27 LKFS dialog-gated, -2 dBTP.
Every row, with its source and a gain calculator, is on the LUFS loudness targets chart.
How Valmera Normalizes Loudness
In Valmera, mastering is one request: “master it,” or “the export sounds quiet.” The agent turns on master loudness, which normalizes the final mix to -14 LUFS integrated with a -2 dBTP true-peak target and a hard ceiling. It applies identically to the preview and the export, so what you approve is what ships, and it stays on through later edits.
It is one target for social and streaming, not a dial: for a -16 LKFS podcast or a -23/-24 broadcast deliverable, master the exported audio in a dedicated tool. It changes overall loudness, not balance; music levels and ducking handle that, as the audio and music docs explain.
Related Terms
- Sidechain ducking: music that dips under a voice and swells back in the gaps.
- Silence detection: the thresholds that decide what counts as a pause.
- Edit decision list: where a mastering setting lives, so every render applies it.
Frequently Asked Questions
Made with Valmera: real exports
Master Your Video Without Opening a Meter
Upload your footage, say “master it,” and approve a preview normalized exactly like the export. Creating an account and uploading are free. Editing requires a subscription.
Create an account →