What this does
It measures how loud a recording is, the way streaming services and broadcasters do, and sets it to the loudness you need: -14 LUFS for music streaming, -16 for podcasts, -23 or -24 for broadcast, or a figure of your own. It measures the true peak as well, the highest point of the actual waveform, and never raises the level past a ceiling, so the result does not clip when it is encoded or played.
The whole job runs in your browser. Your audio is never uploaded, which matters for the episode that is not out yet.
How to use it
- Drop an audio file onto the tool, or press “Choose an audio file”. Its loudness and true peak appear once it has been measured.
- Pick a target, a peak ceiling, and 16-bit or 24-bit WAV. The line above the button says exactly what will happen: the gain, and where the loudness and the peak will land.
- Press “Normalise”, listen to the result, and download it.
Measured against the standard
Loudness meters are checked against the EBU’s Tech 3341 test signals, whose correct readings are published, so a meter can be verified rather than trusted. These are this tool’s readings, from its own test suite:
| EBU Tech 3341 signal | Correct reading | This tool |
|---|---|---|
| Case 1: steady tone at -23 dBFS | -23.0 LUFS | -22.99 |
| Case 2: steady tone at -33 dBFS | -33.0 LUFS | -32.99 |
| Case 3: quiet stretches either side, gated relatively | -23.0 LUFS | -23.01 |
| Case 4: near-silence either side, gated absolutely | -23.0 LUFS | -23.01 |
| Case 5: louder and quieter stretches | -23.0 LUFS | -22.98 |
The standard allows 0.1 LU either way; every reading is within 0.02. The test suite also checks the same tone at 44.1 kHz, which reads -22.99, and checks the ear-weighting filters against the exact coefficients the standard publishes.
True peak is checked with a tone placed so that every sample lands at 71% of the wave’s real height: the samples read -9.03 dB, the wave peaks at -6.02, and the meter reads -6.02. The EBU allows a meter of this kind 0.4 dB under to 0.2 dB over. Tones right across the audible range, from 100 Hz to 20 kHz, read within 0.01 dB under and 0.03 dB over, which matters because true peak is the figure that sets the ceiling. The browser test also measures the tool’s own results with FFmpeg’s independent meter, and the two agree. Why the peak between samples matters, and how far a meter can miss it, is measured in True peak vs sample peak, and how far a meter can be wrong.
Why it stops at the peak ceiling
Raising the level raises everything, peaks included. A quiet recording with sharp peaks, a laugh or a slammed consonant, can run out of headroom before it reaches the target. The tool then raises it only until the true peak touches the ceiling, and tells you where the loudness ended up.
Some recordings already peak above the ceiling: a quiet piano piece with one hard chord, say, or a file mastered right up to full scale. Those cannot be raised at all, so the tool turns them down just enough to sit under the ceiling, even when the target is louder, and says so before you press anything. The same goes for a loud file whose peaks would still be over the ceiling at the target: it comes down a little further than the target asks.
A limiter could squeeze those peaks down to make room, and some normalisers apply one silently. That changes the sound, sometimes audibly, so this tool leaves it to you: it says how far short of the target the file landed, and an editor’s limiter can take it the rest of the way if you want it to.
What happens to the file
Only the level changes: one gain, applied to every sample alike. The file is decoded at 48 kHz, the rate the loudness standard is written for and the standard rate for video, and written as WAV. At 16 bits, triangular dither is added before rounding, so quiet fades stay smooth rather than turning gritty; at 24 bits there is room enough not to need it.
Limitations
WAV out only. Most platforms accept WAV, but if yours wants MP3 or M4A, convert the normalised file in your editor or encoder afterwards.
Long files need memory. The whole recording is decoded in your browser. An hour of stereo takes over a gigabyte, which a laptop handles and an older phone may not.
Video files are not supported. Extract the audio track first.
Mono is measured as the standard says. A mono file is measured as one channel, so it reads 3 LU lower than the same sound in stereo, which is correct by BS.1770 but worth knowing if your platform treats mono files as stereo.