Why -14 LUFS is the New 0 dB

In 1992, 0 dBFS sample-peak was the only ceiling that mattered. RMS meters showed -8 to -6 dB RMS, limiters shaved transients, and whoever was loudest won the A/B in the car. By 2008, some commercial CDs hit -6.5 dB RMS with crest factor under 5 dB. It measured loud. It sounded small.

1. The loudness war was a measurement problem

RMS does not hear like humans. A 100 Hz sine and a 3 kHz sine at same RMS are not equally loud. Our ears follow something closer to K-weighting — defined in ITU-R BS.1770-4: a high-shelf at ~1.68 kHz +4 dB, Q 0.707, and a high-pass at 38.135 Hz. That weighting predicts loudness across spectrum far better than RMS or flat-weighted power.

Worse, RMS has no gating. Intro silence, breaks, and fade-outs drag the number down, so engineers pushed everything up. The result: hyper-compressed choruses, dead verses, no LRA.

2. EBU R128 fixed broadcast, then streaming copied it

EBU R128 (2011) did three things that killed the war:

• Target: -23.0 LUFS ±0.5 LU integrated for broadcast program, not peak.
• Two-stage gating: absolute gate at -70 LUFS excludes silence, relative gate at -10 LU below absolute-gated loudness excludes quiet background that would skew the measurement. This is why EBU R128 in 5 minutes matters — without gating, a podcast with long pauses measures too quiet.
• True-peak ceiling: -1.0 dBTP measured with at least 4× oversampling (48 kHz → 192 kHz) to catch inter-sample overs. Read true peak vs sample peak and dBTP explained.

Broadcast adopted -23 fast. Streaming needed louder for headphones and noisy environments, but wanted the same principle: normalize everything to one reference so users stop reaching for volume.

3. What -14 LUFS actually measures

Integrated is the whole-program number with both gates applied. Algorithm per ITU:

1. Apply K-weighting to each channel, apply channel weights (L/R 1.0, C 1.0, Ls/Rs 1.41 for 5.1).
2. Slice into 400 ms blocks, 75% overlap → compute mean-square.
3. Exclude blocks below -70 LUFS absolute.
4. Compute integrated loudness of remaining blocks, then exclude blocks < integrated -10 LU (relative gate).
5. Re-integrate survivors → your integrated LUFS.

That is what JSM Loudness implements exactly, verified against EBU reference file: -14.05 LUFS integrated, TP -3.08 dBTP, Short-term -11.88, Momentary -11.23, LRA 3.30, PLR 10.97. No approximation, same biquad coefficients: Stage1 1681.97445 Hz +3.999843 dB Q0.707175, Stage2 38.13547 Hz Q0.500327.

What is LUFS? explains momentary (400 ms) and short-term (3 s) too — short-term tells you if your chorus parks above -11 for too long and will feel crushed after normalization.

4. Normalization: why louder is now smaller

Say you master to -8 LUFS integrated, true-peak -0.1 dBFS sample-peak. Spotify Normal applies -6 dB gain. YouTube does same. Result:

• Perceived loudness = identical to a -14 LUFS track.
• But your true-peak now sits at -7.1 dB after gain? No — the gain is applied after decoding, and your inter-sample peaks that were at +0.7 dBTP now become -5.3 dBTP, but you already distorted the DAC and codecs.
• Your PLR is 8 -? Actually -0.1 - (-8) = 7.9 dB. A dynamic -14 LUFS master with TP -1.0 gives PLR 13 dB. After normalization, the dynamic track hits harder because transients survive.

Listeners describe it as: loud masters sound thinner and more fatiguing on normalized playlists. Meters confirm: PSR (Peak-to-Short-term) 5 dB vs 8 dB.

5. True-peak and codec headroom

0 dBFS sample-peak ≠ 0 dBTP. Reconstruction filter overshoot adds up to +0.6 dB on real DACs. MP3/AAC encoders need even more — 320k MP3 can overshoot +1.2 dB. That is why -1.0 dBTP is the new 0 dB. If you master to -0.3 dBTP and platform transcodes to AAC 256k, you create audible clipping on phones. Deep dive on DBTP has codec tests.

Fix: limiter ceiling -1.0 dBTP true-peak, 4× minimum, 8× or 16× better in final check. Your ISP meter must oversample.

6. How to work at -14 today

Workflow in JSM Loudness (100% browser, Quebec Law 25 compliant — see privacy note):

1. Drop WAV/AAC/MP3 — nothing leaves your device.
2. Check integrated. If -11.2, you need -2.8 dB gain or less limiting.
3. Check short-term graph: if chorus sits at -9 short-term for 40 seconds, pull 2 dB.
4. Check true-peak. If -0.4 dBTP, lower limiter ceiling to -1.0 dBTP true-peak.
5. Export gain-adjusted file, re-check. PLR should rise.

7. When to break -14 on purpose

Three cases:
Club / DJ: you still deliver -9 to -7 LUFS integrated for raw files used on CDJs without normalization, but keep a -14 streaming version.
Short-form: TikTok measures heavily on 3-second short-term; some creators master vertical hooks to -11/-9 short-term while integrated stays -14.
Intentionally crushed aesthetic: lo-fi, hyperpop. Do it knowing it will be turned down.

In all cases, never go above -1.0 dBTP true-peak. That rule has no exception.

Bottom line: -14 LUFS is not a limiter preset. It is a listening reference, like 83 dB SPL in film. When everyone aims there, tracks with dynamics win A/Bs. The war ended because measurement finally matched hearing. Stop chasing 0 dBFS sample peak. Start watching integrated LUFS, true-peak dBTP, and PLR. Those three decide how you translate on every phone, speaker, and codec.