LUFS · TRUE PEAK · ONE GAIN
Drop your audio file here or click to browse
MP3, WAV, FLAC, M4A, AAC, OGG, Opus · one file, up to 20 minutes
Loudness stopped being a matter of taste the day streaming platforms started turning everything down to the same level. Upload a master at −8 LUFS and Spotify plays it at −14 anyway; all the loudness war bought was a flatter record played at the same volume as everyone else's. What matters now is knowing the number before you deliver, and moving it by the smallest change that gets you there.
This page measures the number the way the platforms do — ITU-R BS.1770-4, the same standard behind EBU R128 — and then applies a single gain to reach your target. Nothing is compressed and nothing is limited. Everything happens on your own machine; the file is never uploaded.
| Where it is going | Target | Why |
|---|---|---|
| Spotify, YouTube, Tidal, Amazon | −14 LUFS | They normalise to about this, so anything louder is turned down on the way in |
| Apple Music | −16 LUFS | Sound Check targets a little quieter than the rest |
| Podcasts and spoken word | −16 LUFS | Mono speech is often −19; most directories publish −16 as the safe figure |
| European broadcast | −23 LUFS | EBU R128, which is a legal requirement in several countries |
There are two different jobs that both get called "normalising". One is making a file measure a particular number, which is a single multiplication and changes nothing about how the audio behaves. The other is making a file feel louder, which needs compression or limiting and genuinely alters the recording. This page does the first, deliberately, and shows the loudness range before and after so you can confirm it: the two figures should be identical.
That also explains the one case where it will not reach the target. If the file already peaks near full scale, the gain needed to raise its loudness would push those peaks past the ceiling. The choice at that moment is to limit the peaks — hitting the number by changing the sound — or to stop at the ceiling and say so. This stops, and reports the shortfall in decibels, because a master that quietly had its transients shaved is worse than one that arrives a decibel and a half quiet.
The implementation is checked against ffmpeg's ebur128 filter — an independent implementation of the same standard — on sine tones at three frequencies, on noise, on mono, stereo and single-channel material, at both 44.1 and 48 kHz, and on a signal built specifically to exercise the gating. Integrated loudness agrees to within 0.04 LU, well inside the ±0.1 LU that EBU Tech 3341 sets for compliance, and the loudness range agrees exactly. That cross-check is worth more than any amount of internal consistency: a loudness meter that is wrong in a self-consistent way looks exactly like one that is right until you compare it with the rest of the world.
LUFS is loudness measured the way people hear it rather than the way a meter does. Plain dB tells you the electrical level; it does not know that the ear is much less sensitive to a 50 Hz rumble than to a voice at 2 kHz, so two files at the same peak level can sound nothing alike. LUFS filters the audio to approximate that sensitivity, averages the result over the whole piece, and throws away the near-silence so a quiet intro does not drag the number down. One LU is one dB, so the scale is familiar even though what it counts is not.
Streaming platforms normalise to roughly −14 LUFS, so that is the useful target for music going to Spotify, YouTube or Tidal: master louder than that and the platform simply turns it down, and you have given away dynamic range for nothing. Apple Music sits nearer −16, and so do most podcast specifications, because spoken word is listened to in cars and on trains where a wide range is a liability. Broadcast in Europe is −23 LUFS under EBU R128. If you do not know where the file is going, −16 is the safest middle.
No. This applies a single gain to the whole file — every sample multiplied by the same number — so the difference between the quietest and loudest passage is exactly what it was. That is why the loudness range is shown before and after and does not move. A compressor or a limiter would change it, and would be the wrong tool for hitting a delivery target: the platform is going to adjust the level anyway, and all a squashed master achieves is arriving with less to work with.
Between any two samples the waveform passes through values no sample recorded, and a converter or a lossy encoder reconstructs those. So a file whose samples all sit at exactly 0 dBFS can still overshoot when it is played or turned into an MP3 — which is where distortion that was not audible in the editor comes from. True peak estimates that overshoot by looking between the samples. Leaving a decibel of headroom is the usual insurance, and it is what most delivery specifications ask for.
Because the gain needed would have pushed the peaks past the ceiling. When that happens the gain is reduced so the peaks land exactly on the ceiling instead, and the shortfall is reported in dB rather than hidden. The alternative — limiting the peaks so the number can be hit anyway — would change how the file sounds while reporting success, and that is the opposite of what someone normalising a finished master wants. If you need both the target and the ceiling, the fix is in the mix, not in the meter.
No. The file is decoded, measured, adjusted and encoded by your own browser, and nothing is sent to a server at any point. You can check it: load the page, disconnect from the internet, and it still works from end to end.