Normalize audio online

Normalizing means bringing a recording to a known loudness instead of whatever level it happened to be captured at. CutNoodle does it in a browser tab, on a file that never leaves your computer, and takes about as long as the file takes to re-encode.

What normalizing actually changes

A recording has no inherent volume. What it has is a level, decided by how close the microphone was, how the input gain was set, and how loudly you happened to speak that day. Two files recorded in the same room an hour apart can land several decibels apart for no reason anybody intended. Normalizing removes that accident: it measures the recording, works out how far it sits from a target, and moves the whole thing by that amount.

The important word is whole. Normalizing applies one offset to the entire file. It does not go through your recording deciding that this sentence is too quiet and that one is too loud. That is compression, it is a different tool, and it is not what this does.

Peak and perceived loudness are not the same thing

This is the distinction that decides whether normalizing gives you what you wanted, and almost every tool that calls itself a normalizer picks one of the two.

Peak normalization looks for the single loudest instant in the file, a door closing, a plosive on the microphone, and scales everything so that instant lands just under the maximum. It is trivial to compute and it is almost useless for speech. One stray thump decides the level of your entire episode, and a recording whose loudest moment is a cough ends up quieter than one recorded at the same volume without the cough.

Loudness normalization measures how loud the recording actually sounds to a person, averaged across its length, and moves it to a target expressed in LUFS. Momentary peaks stop being in charge. Two files normalized this way sit at the same apparent volume even if their peaks are nowhere near each other, which is the thing you wanted when you asked for the volume to be evened out.

CutNoodle does the second kind, with a true peak ceiling so the first kind cannot cause damage on the way. If you want to know what the LUFS numbers mean and which one your platform expects, how loud should a podcast be is the reference page for that; nothing here requires you to read it first.

Doing it

• Open the tool. The Simple Audio Editor runs in this browser tab. Choose a file and it is ready at once, because there is nothing to upload: MP3, WAV, M4A, AAC, OGG and FLAC all open.
• Trim first, if you need to. Split at the playhead and delete the countdown at the start and the fumbling for the stop button at the end. Do this before normalizing rather than after: dead air at the ends is part of what gets measured.
• Switch on Normalize volume. One checkbox. There is no target to choose, because the target for spoken word is not really a matter of taste and picking it badly is the most common way to get a worse file than you started with.
• Add fades if the ends are abrupt. Select a clip and set its fade in or fade out, in tenths of a second, up to half the clip's length. A half-second fade in on the first clip is usually enough to stop a recording starting on a click.
• Export. MP3, M4A or WAV. The whole chain runs as a single ffmpeg pass on your machine, so a long file is real work for your laptop rather than something a server absorbs for you.

What to expect from the result

A spoken recording that was too quiet comes back at a comfortable listening level, and a file that was recorded hot comes down to meet it. Played one after the other, two files you normalized will match. That is the whole promise, and it is worth being clear that it is the whole promise.

What will not change: the noise floor moves with the voice. Raising a recording by twelve decibels raises the air conditioner by twelve decibels too, so a hissy file comes back as a louder hissy file. Nothing here removes noise, and nothing here repairs a recording. Why your audio is too quiet goes through what is recoverable and what is not, which is worth reading before you spend an evening trying.

The other thing that does not change is dynamic range. If you moved toward the microphone during an anecdote and away again afterwards, those passages stay as far apart as they were. Normalizing sets where the recording sits; it does not flatten what happens inside it.

Common questions

CutNoodle is free and runs in your browser. Open the Simple Audio Editor and pick a file.