Audio Normalizer
Hit the exact loudness your platform is measuring.
or paste one from your clipboard
| Measurement | Value | What it tells you |
|---|---|---|
| Integrated loudness | · | How loud the whole file sounds |
| Loudness range | · | How much the level moves across it |
| True peak | · | The highest point after reconstruction |
| Sample peak | · | The largest value actually stored |
How to use it
- Drop your file onto the page. It is measured immediately with the ITU-R BS.1770-4 algorithm, and the panel fills in with integrated loudness, loudness range and true peak.
- Choose the platform you are delivering to, or set a target in LUFS by hand if you are working to a spec of your own.
- Leave the true-peak ceiling at −1 dBTP unless you have a reason not to. It protects against inter-sample peaks that only appear once the file has been encoded.
- Read the summary line. It states the exact gain about to be applied, and says so plainly when the ceiling stops that gain short of your target.
- Download. Normalizing applies one static gain to the whole file, so the balance and dynamics of the recording are untouched.
What integrated loudness measures
LUFS stands for Loudness Units relative to Full Scale, and the figure this tool reports is an integrated one: a single number for the whole programme, first sample to last. It comes from the ITU-R BS.1770-4 algorithm, which does three things an ordinary level meter does not. It applies K-weighting, a filter pair that trims deep bass and lifts the presence region, because a 40 Hz rumble and a 3 kHz voice at the same electrical level are nowhere near equally loud to a person. It works in overlapping 400-millisecond blocks rather than instant by instant. And it gates: blocks below −70 LUFS are thrown out, and then blocks more than 10 LU below the average of what remains are thrown out too, so pauses between sentences and the fade at the end do not drag the number down.
That gating is why integrated loudness tracks what people report hearing, and why it replaced peak normalization everywhere it mattered. A peak meter describes one sample (the single loudest instant in a file of forty million) and says nothing about the rest. Two tracks can both peak at −0.1 dBFS while one is a heavily limited pop master and the other a solo piano recording, and still differ by 12 LU in how loud they sound. Once services began playing material from thousands of sources back to back, matching peaks stopped being useful.
Why delivering louder than the target buys nothing
Every major platform measures your file on ingest and applies a static gain at playback so it lands on their target. Deliver a master at −8 LUFS to a service targeting −14 and it does not play back 6 dB louder than anyone else; it plays back at −14, with 6 dB of attenuation applied. Everything you gave up to reach −8, the transients you limited away, the dynamic range you compressed out. Is gone, and it bought you nothing. You are left with a flatter, more fatiguing version of your own mix, playing at the same volume as the version you did not squash.
The reverse is safer than most people expect. Platforms turn quiet material up as well as loud material down, though several will decline to raise a track when doing so would push its true peak past their ceiling. That is exactly the case this tool reports: when the ceiling stops the gain short, the file lands above target on purpose, and the platform finishes the job at playback.
True peak versus sample peak
A digital file stores samples, not a waveform. The waveform is what a converter reconstructs between those samples on the way to a speaker, and that reconstruction can rise above the highest sample it was built from. A file whose largest stored value reads −0.2 dBFS can reach +0.4 dBTP when played, and a lossy encoder pushes it further still, because MP3 and AAC do not decode back to precisely the samples they were handed. This tool oversamples four times to estimate the peak between samples. Holding the ceiling at −1 dBTP costs a decibel nobody will ever hear and prevents the crunch that otherwise shows up only after encoding, on someone else's playback chain.