What a LUFS meter measures, in five steps

Streaming services, broadcasters and podcast platforms all talk about loudness in LUFS. The number looks like magic, but the recipe is short and public: ITU-R BS.1770-4. This is that recipe in five steps, as seance's meter runs it, with the tests that keep it honest.

1. Weight the frequencies like an ear#

Ears are not equally sensitive to all frequencies. The meter first runs the audio through the K-weighting filter: a gentle boost of about 4 dB above roughly 1.7 kHz, where we hear best, and a high-pass that removes most of what lies below about 40 Hz. Deep bass that you feel more than hear counts for little.

The standard gives the two filters as coefficients for 48 kHz. seance designs them afresh for your file's own sample rate from their analogue shapes, so a 44.1 kHz file is weighted the same way and not by coefficients made for another rate.

2. Measure energy in short blocks#

The filtered signal is squared and averaged in blocks of 400 ms, each overlapping the last by three quarters, so a new block starts every 100 ms. For stereo, the two channels' energies are added. (Surround channels get their own weights in the standard; seance weights every channel the same, which is right for mono and stereo.)

Each block becomes a loudness with one formula: −0.691 + 10 × log₁₀ of its mean square. That −0.691 is there so that a 997 Hz tone comes out at the same number with and without the filter.

3. Ignore the silence#

If a song ends with ten seconds of silence, it should not count as quieter. So the meter gates:

• Every block quieter than −70 LUFS is thrown away. That is the absolute gate.
• The average of the rest is taken, and every block more than 10 LU below that average is thrown away too. That is the relative gate.

The integrated loudness is the average energy of the blocks that are left, turned into LUFS the same way. It is an average of energy, not of LUFS values, so loud passages weigh more than quiet ones.

4. Short-term, momentary and range#

• Momentary is one 400 ms block. The meter reports the loudest.
• Short-term is a 3-second window, moved every 100 ms. Again the meter reports the loudest, and seance draws the whole series against your target.
• Loudness range, from EBU R128, looks at the short-term values: gate at −70 LUFS, gate again 20 LU below their average, then take the spread between the 10th and the 95th percentile. A heavily compressed master shows a small range, a film mix a large one.

5. Peaks between the samples#

A sample peak is the largest sample value. But a digital-to-analogue converter does not play samples, it draws a smooth wave through them, and that wave can rise above every sample. A true-peak meter estimates that wave by oversampling: seance computes three extra points between every two samples with a 48-tap interpolation filter, four times the original rate, and reports the highest.

The tests that keep it honest#

A loudness meter that is a few tenths off looks exactly like one that is right. So the audio suite feeds it signals whose answer is known, on every build:

Signal	Why the answer is known	seance reads
997 Hz sine at full scale, one channel	its mean square is ½, which is −3.01 dB, and the filter and the −0.691 cancel there	−3.01 LUFS
The same sine 20 dB lower	20 dB lower	−23.01 LUFS
The full-scale sine	its peak is exactly 1.0	0.00 dBTP
A sine at a quarter of the sample rate, shifted by 45°	every sample lands at 0.707 (−3.01 dB) while the wave reaches 1.0	−0.07 dBTP
2 s of that sine at −6 dB, then 8 s at −80 dB	the gate must drop the quiet 8 s, which leaves −9.03	−9.37 LUFS; the blocks that straddle the change still count

What the number is for#

Platforms such as Spotify and YouTube turn playback up or down so that songs sit at a similar loudness, and they use a measure of this family to decide. Knowing your integrated loudness tells you how much they will change your master. Knowing your true peak tells you whether a lossy encoder, which also rebuilds the wave, is likely to clip it. Neither number tells you whether it sounds good.

You can measure a file in your browser with seance's meter. Nothing is uploaded.