The loudness war peaked around 2008 to 2010, when commercial pop and rock masters routinely measured -6 to -7 LUFS integrated. Metallica's Death Magnetic, released in 2008, became infamous for distortion audible even on CD, with fans circulating a petition for a re-master. Then streaming platforms standardized their normalization targets, and within a few years, mastering engineers were quoting -14 LUFS like a mantra.
But few people understand what that number actually measures. LUFS is not a volume knob. It is the output of a specific DSP algorithm, and the details of that algorithm explain some counterintuitive behavior: why a dynamically dead master sounds no louder than a well-preserved one on Spotify, why your fade-outs don't drag down your loudness measurement, and why a master that peaks at exactly 0 dBFS can still audibly clip after encoding.
Where LUFS Comes From
LUFS stands for Loudness Units relative to Full Scale. The algorithm that produces it is defined in ITU-R BS.1770, a standard published by the International Telecommunication Union in 2006, now on its fifth revision (BS.1770-5, November 2023). EBU R128, published by the European Broadcasting Union in August 2010, built a complete broadcast loudness standard on top of BS.1770, specifying a target of -23 LUFS integrated, a tolerance of plus or minus 0.5 LU, and a maximum true peak of -1 dBTP.
The LUFS measurement pipeline has four stages: K-weight, square, gate, integrate.
Stage One: K-Weighting
Before any averaging, the signal passes through the K-weighting filter. This is a two-stage IIR filter designed to model how the human ear actually perceives loudness at different frequencies.
Stage one is a high-frequency shelving filter that adds approximately +4 dB above roughly 1.6 kHz. The justification is acoustic: your head acts as a rigid baffle, and sound arriving at your ear from a speaker in front of you is slightly boosted in the high-mid range compared to a microphone measurement in free space. The standard explicitly models the head as a rigid sphere to derive this pre-filter.
Stage two is a high-pass filter with a -3 dB point near 38 Hz. Sub-bass content below the threshold of audibility should not contribute to perceived loudness.
The combined effect is a frequency-weighted signal that better reflects what a listener actually hears. A mix with heavy sub content gets measured closer to how it feels rather than how its raw energy would suggest.
Stage Two: Squaring and Averaging
After K-weighting, the algorithm computes mean square power over overlapping 400 ms blocks. This is the basis for momentary loudness, the reading on the fastest meter in your DAW. Short-term loudness averages a 3-second window. Integrated loudness, the single number that streaming platforms care about, averages the blocks over the full duration of the track, but with a critical twist.
Stage Three: Gating
Two gates are applied in sequence, and this is where LUFS diverges significantly from a simple RMS average.
The absolute gate discards any 400 ms block that measures below -70 LUFS. True silence and noise floor are eliminated.
The relative gate is more consequential. First, the algorithm computes the absolute-gated average across all non-silent blocks. Then it sets a threshold 10 LU below that average and recomputes, keeping only the blocks that exceed this second threshold. Quiet passages, long ambient intros, slow fades, and sparse sections all get excluded if they fall below the relative gate.
This means your three-minute electronic track with a 30-second ambient intro is not penalized for that intro. The intro blocks fall below the relative gate, get dropped from the calculation, and the integrated loudness reflects the body of the track where the music actually lives.
It also means you cannot game the measurement by padding a loud master with silence. The gating algorithm removes those silent blocks immediately.
The Streaming Targets
EBU R128 targets -23 LUFS for broadcast, calibrated for the listening environments and monitoring levels typical in European broadcasting. Streaming platforms chose louder targets, calibrated for earbuds and phone speakers in noisy environments:
- Spotify: -14 LUFS integrated, -1 dBTP true peak limit
- Apple Music: -16 LUFS integrated, -1 dBTP true peak limit
- YouTube: -14 LUFS (attenuation only, does not boost quiet masters)
- Amazon Music and Tidal: -14 LUFS
- Deezer: -15 LUFS
Each platform measures integrated loudness after upload and applies a gain offset at playback. A track at -18 LUFS gets boosted 4 LU on Spotify. A track at -8 LUFS gets attenuated 6 LU.
The implication is direct: mastering louder than your target platform's normalization point yields no loudness advantage. A master at -8 LUFS plays back on Spotify at the same perceived loudness as one at -14 LUFS, but the -8 LUFS version went through heavy limiting to get there, and that limiting is audible after the attenuation is applied. The gain reduction artifacts, previously masked by the sheer volume, become clear when the playback system turns the track down 6 LU.
Not all platforms boost. YouTube, Tidal, and Amazon only attenuate loud tracks. A quiet master on these services plays back at exactly its native loudness. Whether you treat this as a reason to master louder for those platforms is a trade-off worth knowing about.
True Peak: The Problem That Sample Peak Meters Hide
A digital audio file stores discrete samples. Your DAW's peak meter shows the highest sample value. But when a DAC reconstructs those samples into an analog waveform, the actual signal between samples can exceed any individual sample value. These inter-sample peaks exist only in the reconstructed waveform. They are invisible to a standard sample peak meter.
They get worse after codec transcoding. When Spotify encodes your WAV to Ogg Vorbis, or Apple Music converts to AAC, the mathematical operations of lossy compression can push peaks significantly above the original level. A master that reads exactly 0 dBFS on your peak meter may produce inter-sample peaks of +2 dBTP or more after encoding, causing audible clipping on the listener's DAC.
This is the specific reason the -1 dBTP ceiling exists in EBU R128 and across streaming platform specifications. It provides headroom for analog waveform reconstruction and for the peak-shifting behavior of codec encoding. A hard clip at 0 dBFS on your sample meter does not equal 0 dBTP on the reconstructed waveform.
True peak meters oversample the signal (typically 4x or 8x) to estimate where the reconstructed waveform's maximum actually falls. When a limiter has true peak detection enabled and its ceiling is set to -1 dBTP, it is constraining the oversampled reconstruction, not the sample values. This is meaningfully different from setting a limiter ceiling to -1 dBFS on a standard sample peak meter.
Spotify, Apple Music, and EBU R128 all specify -1 dBTP. The practical recommendation: use a limiter with true peak detection, set the ceiling to -1 dBTP, and verify with a true peak meter rather than a sample peak meter.
What the Loudness War Actually Taught Us
The loudness war did not end because mastering engineers collectively decided to use less limiting. It ended because streaming infrastructure made loudness a zero-sum game. The playback normalization guarantee means that any integration score above the platform target is simply attenuated back down, taking all the dynamic compression along with it.
The K-weighting filter, the gating algorithm, and the platform targets are not arbitrary rules. They represent a calibrated attempt to measure what listeners actually perceive, and the codec headroom requirement is a practical acknowledgment of how audio survives digital distribution. Understanding each stage of the pipeline is more useful than memorizing the target numbers.
-14 LUFS is a consequence of the math, not a goal in itself.