My audio came out too quiet

Audio usually comes out too quiet because the microphone was too far away, the input gain was set too low, or the file was never normalized to a standard loudness. A clean but quiet recording can be fixed in seconds by normalizing it. A recording where the voice sits close to the background noise cannot be rescued, because raising the level raises the noise by the same amount.

A recording that plays back faint is one of two problems wearing the same coat. One of them takes half a minute to fix and the other cannot be fixed at all. This is how to tell which one you have before you spend an evening on it.

Why a file comes out quiet

The microphone was too far away. Level falls off steeply with distance: doubling the gap between your mouth and the capsule costs you about six decibels, and a laptop microphone across the desk is several doublings away from where a headset sits. This is the most common cause and the most consequential, because distance does not only reduce the voice. It also brings the room up relative to it, so you lose signal and gain noise at the same time.

The input gain was set low. Every interface, mixer and recording app has a level control, and the sensible habit of leaving headroom so nothing clips becomes a problem when it is overdone. A recording that peaks at -30 dBFS was captured with the gain far lower than it needed to be. On its own this is the benign case: the recording is clean, it is just small.

Nothing ever set the level. Most recording apps hand you the raw capture. Nothing in the chain measures how loud the result is or moves it toward a standard, so whatever the first two factors produced is what you get. Publishing platforms assume somebody did that step; if nobody did, your file arrives at whatever level it happened to land on.

Those three stack. A phone on a table, recorded with conservative gain, never levelled afterwards, is a file three separate decisions made quiet.

Why turning up the player is not the fix

It fixes it for you, on that device, for that listen. The file is unchanged, so the next person to open it is back where you started, and every platform that measures what you upload still sees a quiet file.

It also fails in a way that is easy to miss. Podcast apps, YouTube and Spotify all measure incoming audio and adjust playback toward their own target. If your file arrives well under that target, it gets turned up for the listener automatically, along with the hiss, and you have no say in how. The volume slider you used while checking told you nothing about any of this, because it was applied after the file left the equation.

Changing the file is what makes the result the same for everyone. That is a measurement and a single offset, which takes seconds: normalize audio online walks through it, and how loud should a podcast be explains what the platforms are measuring against.

What comes back and what does not

The question that decides everything is not how quiet the voice is. It is how far the voice sits above the noise underneath it, because raising one raises the other by exactly the same amount. A clean recording at -30 dBFS is a full recording that happens to be small, and bringing it up gives you a full recording at a normal level. There is nothing lossy about it.

The failure case sounds identical until you fix it. Take a voice recorded far from the microphone in a room with a fan: the voice is low and the fan is not far below it. Raise the file by eighteen decibels and you get a normal-sounding voice over a fan that has become impossible to ignore. Nothing was damaged, but nothing was rescued either. You have simply made the recording loud enough to hear what was wrong with it.

The other one-way door is clipping. If the level was too high rather than too low and the waveform hit the ceiling, the peaks were discarded at the moment of recording and no amount of processing invents them back. This is worth knowing because it is the risk on the other side of setting gain generously, and it is why conservative gain is the more common mistake: too quiet is usually recoverable, too loud is not.

Be clear-eyed about the limit here. Besides cutting and fading, CutNoodle changes the level of a recording and nothing else. It has no noise reduction, no equaliser, no gate and no de-esser. Normalizing does not save a recording that was made badly, and no honest tool will tell you otherwise. What it does is stop a recording that was made fine from arriving quiet.

Doing it better next time

• Get closer. A hand's width from the microphone beats any processing you can apply afterwards, and it costs nothing.
• Watch the meter while you record. Peaks landing around -12 to -6 dBFS leave room for a surprise without burying the voice. If the meter barely moves, the take is already compromised.
• Record a few seconds of silence. It tells you where your noise floor sits, which is the number that decides whether a quiet take is recoverable.
• Level the file before publishing. It is the last step, it takes seconds, and it is the one nobody does.

Common questions

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