Audio Normalizer

Bringing the level up

1. Drop the quiet file on the Audio file box. It decodes and gets a player of its own underneath, which is worth thirty seconds of your attention, because whatever is loudest in there decides how far everything else can rise.
2. Type a target peak. 100 leaves nothing spare, 89 keeps about a decibel back, and either way the figure applies to the file from start to finish.
3. Pick M4A, WAV, Opus or FLAC, then press Normalize & download. A progress bar runs during the encode with Cancel beside it, and what comes back is a player plus a Download button, the file named after the one you loaded.

About this tool

A recording that sounds fine on your headphones and vanishes on someone else's phone usually has one thing wrong with it: its loudest moment sits well below full scale. Peak normalizing is the fix. The tool finds the single biggest sample in the file, works out what it would take to lift that one sample to your target, then multiplies every sample by the same number. Nothing about the balance of the recording shifts. All of it moves together.

Target peak is a percentage box that steps in whole numbers between 1 and 100. It starts at 100, which puts the loudest sample right at 0 dBFS. When you want headroom instead, the conversions are worth memorising: 89 is about a decibel down, 79 is two, 71 is three, 50 is a clean six. I leave it on 89 for anything heading back into an edit, since a file pinned at full scale has nowhere to go once the next tool adds gain.

Output goes to M4A, Opus, FLAC or WAV, and the WAV writer works at 16-bit, keeping whatever sample rate the source had. The other three get encoded from that same 16-bit copy, which makes this a good step for voice notes, lecture audio, podcast rough cuts and samples, and the wrong step for finishing a master. Matching several clips is a matter of running each one at the same target. Recorded in the Voice Recorder, trimmed in the Audio Cutter, normalized here: that is the order I use.

Good to know

• One loud moment decides the whole file. A cough, a door, a knuckle on the desk: whichever sample is biggest gets lifted to your target, and the speech underneath rises by that same modest amount. Cut the spike out in the Audio Cutter and run it again to see how much difference it makes.
• Peak is a different quantity from loudness. Two files normalized to 89 can still feel unequal, because ears average energy over time while a peak meter reports one instant. Streaming platforms measure LUFS, and this tool has nothing to say about landing on -14.
• Dynamics come out exactly as they went in. There is no compressor and no limiter here, so the gap between the quietest passage and the loudest one is the same gap it always was.
• Distortion already baked into a recording stays baked in. Lifting the level carries the damage up with everything else.
• Every export is written at 16-bit, FLAC included, and a lossy source picks up one more encode on the way out. That is fine for a voice memo. It is the wrong place to prepare audio for a mastering engineer.

Frequently asked questions