Loudness

EBU R128, LUFS, and the number your programme has to hit

Every platform asks for a level. This is what the measurement means, where each target comes from, and what Seston does about it.

What LUFS measures

A peak meter says how tall the waveform got. It says nothing about how loud something sounds - a bright, compressed voice and a quiet piano can peak in the same place. LUFS, Loudness Units relative to Full Scale, measures perceived loudness instead. It filters the audio the way an ear weights frequencies and then averages the energy over time.

The method is defined in ITU-R BS.1770, which is what every loudness meter in broadcast implements. One unit, 1 LU, is one decibel, so a programme at −18 LUFS is 2 LU quieter than one at −16.

Three readings are usually quoted.

• Integrated is the whole programme, first sample to last. This is what a platform means when it names a target.
• Short-term is a three-second window, for watching a mix move.
• True peak is not loudness at all. It is the highest level the waveform reaches between samples once it has been reconstructed, measured in dBTP, and it is the reason a file that never exceeded 0 dBFS can still distort after encoding.

What EBU R128 says

EBU R128 is a recommendation from the European Broadcasting Union that tells broadcasters to normalise to programme loudness rather than to peak. It sets the target at −23 LUFS, with a tolerance of ±0.5 LU where a programme can be mixed to it, and a maximum true peak of −1 dBTP.

That one decision is why television in Europe stopped being loud in the adverts and quiet in the drama. It says nothing about podcasts, and −23 LUFS is far too quiet for one.

Which target belongs to what

• Broadcast in Europe, −23 LUFS and −1 dBTP. EBU R128. If you deliver to a station this is usually in the contract.
• Podcast, −16 LUFS. A convention rather than a standard. Apple Podcasts asks for −16, and the AES recommendation for streaming speech sits in the −16 to −20 range. Spoken word carries well there without sounding squashed.
• Music streaming, about −14 LUFS. Spotify and others normalise playback to roughly this and turn louder masters down rather than turning quieter ones up, so mastering above it buys nothing.

The number is half of it. A programme normalised to −16 LUFS with peaks at 0 dBFS will distort on an encoder, which is what the true peak ceiling prevents, and −1 dBTP is the usual answer.

What Seston does

Seston measures to ITU-R BS.1770 and reports what it found before it changes anything. Integrated loudness, true peak, and how far each one is from the target you chose. The export presets carry the three targets above, so Podcast — MP3 128 mono, −16 LUFS is one choice rather than a number to remember.

The correction is a gain change and not compression. A programme that reads −21 LUFS against a −16 target is lifted by 5 LU and nothing else happens to it. The limiter that protects the ceiling looks ahead, so peaks are held down rather than clipped flat, and it engages only where the material needs it.

Where a recording is uneven rather than quiet, a guest twice as loud as the presenter, loudness normalisation is the wrong tool and will not fix it. That is what the compressor and the per-clip volume line are for, and they belong before the measurement rather than after it.

Four mistakes worth avoiding

• Normalising twice. Loudness is measured across the whole programme. Normalising the parts and then the whole moves the parts against each other for no reason.
• Mixing up mono and stereo. A mono file and a stereo file of the same material do not measure the same. Check what the platform means before delivering mono at a stereo target.
• Chasing the peak meter. Getting a voice to touch 0 dBFS is not loudness. It is how a programme ends up loud, thin, and still quieter than the one before it.
• Ignoring true peak. A file that measures −16 LUFS and 0.0 dBFS can read +1.4 dBTP after encoding, which is audible distortion on some players.

In short

Measure the whole programme, normalise to the target the platform asks for, leave a decibel of headroom at the top, and fix unevenness before you measure rather than after. Seston does the measuring and the arithmetic, and reports every number it used.

Seston is free for Windows, and the loudness section of the handbook covers the buttons rather than the theory.