Change audio volume

This page multiplies every sample in your audio by the same number. You can give that number directly, give it in decibels, or ask for the loudest sample to be scaled to a level you choose. Before anything is written it tells you where the new peak lands, and if the setting would push audio past what the format can store it says so — because clipping is permanent once saved.

What volume actually is, and why decibels are not a percentage

Every sample in a digital audio file is a number between −1 and 1, describing how far the speaker cone should be from its resting position at that instant. Changing volume means multiplying all of those numbers by the same amount. Multiply by 2 and the waveform is twice as tall; multiply by 0.5 and it is half.

Decibels describe the same thing on a logarithmic scale, because that is closer to how hearing works. Six decibels is roughly a doubling or halving of amplitude — so −6 dB is about ×0.5 and +6 dB is about ×2. Twenty decibels is a factor of ten. This is why a decibel figure is never a percentage: +100% would be +6 dB, and there is no such thing as “200% volume” in any meaningful sense.

Perceived loudness is a third thing again, and it does not track amplitude neatly. Doubling the amplitude of a recording does not make it sound twice as loud; the usual rule of thumb is that a tenfold change in power, about 10 dB, is heard as roughly a doubling. That is why this tool reports the numbers it actually applied — the multiplier, the decibel change, and the old and new peak — rather than claiming a result in terms of how loud it will feel.

Clipping: the one way this operation can destroy audio

A 16-bit sample cannot hold a value beyond the maximum the format defines. When a multiplier pushes samples past that ceiling they are flattened to it, and the smooth curve of the waveform becomes a straight line across the top. That is clipping, it sounds like harsh distortion, and it cannot be undone afterwards — the shape of the original peak is simply gone.

This is what makes “make it louder” more dangerous than it looks. A file whose loudest sample is already at 0.9 has room for about ×1.11 and no more. Push it to ×2 and every peak in the recording is destroyed while the quiet parts merely get louder.

So this page measures the loudest sample in your file when it decodes it, and shows you what your setting will do to it. If the new peak would land above the ceiling, it tells you the level it would reach and the largest multiplier that stays inside range. It does not silently limit or compress the audio to rescue you, because that would be a different processing decision made without asking. Reducing volume, by contrast, is always safe: no sample can leave the range by getting smaller.

Normalising, and why it is not the same as “making it as loud as everyone else”

The normalise option finds the loudest sample in your file and scales everything so that sample lands exactly where you asked, usually −1 dBFS to leave a little headroom. It is one multiplication, applied evenly, and it cannot clip by definition because the peak is the thing being aimed.

What it does not do is match perceived loudness. Two files normalised to the same peak can still sound very different, because one might be a dense, consistently loud mix and the other a sparse recording with one loud cough in it. Matching how loud things feel is loudness normalisation, measured in LUFS, and it is what streaming services use; it needs a different measurement and usually needs to reduce dynamic range as well. This tool does peak normalisation and says so, rather than borrowing the more impressive word.

That distinction has a practical consequence. If your recording has one accidental bang in it, peak normalisation will aim at the bang and leave the speech as quiet as it was. Trim the bang out first, then normalise.

The output, the cost, and the fact that nothing is uploaded

Changing volume means decoding the audio, so the result has to be written back out. Compressing it again would be a second lossy pass, so instead the samples are written as uncompressed 16-bit PCM in a WAV. Nothing is compressed a second time. That is not the same as lossless — if your source was an MP3 or an AAC track, that first encode is not something decoding can undo — but this tool adds nothing new beyond the multiplication you asked for.

Uncompressed audio costs about 10.1 MB per minute at 44.1 kHz in stereo, and the page shows the exact output size before you start. There is a real limit too: 300 MB per file and 30 minutes of audio on a desktop, half on a phone, because the decoded samples have to fit in memory.

No file is uploaded. Open your developer tools, watch the Network tab and run the tool — your audio is in none of the requests, and the Content-Security-Policy header this site sends restricts connections to our own origin, enforced by your browser rather than promised by us. After the first visit the page works with no network at all.