That is what makes normalisation safe. Because one multiplication is applied uniformly to every sample, the operation is reversible and adds no distortion of its own, provided the resulting peak stays at or below full scale. It is the correct first move when an interview was recorded too quietly, or when a set of clips needs to arrive at an editor at a comparable level.
Everything is computed on your machine. The file is decoded to raw samples by the Web Audio API, the peak is found by scanning every value in every channel, the gain is applied through an offline render, and a 16 bit PCM WAV is written out. You can normalize audio online here without a single byte leaving the browser tab.
What an audio normalizer changes and what it leaves alone
Normalisation is a scaling operation, not a shaping one. Every sample is multiplied by the same number, so the ratio between the quietest and the loudest moment is untouched. A recording with a fifty decibel dynamic range still has a fifty decibel dynamic range afterwards, it just occupies a higher part of the available scale.
This is the crucial difference from compression and limiting, both of which change that ratio deliberately. If a podcast still feels uneven after normalising, normalisation is not the missing step. Compression is, and no amount of gain will substitute for it.
- Use it when a whole recording is simply too quiet.
- Use it to bring several separate clips to the same ceiling before an edit.
- Do not use it to fix a recording where some parts are far louder than others.
- Do not use it on material already mastered close to full scale, since there is no headroom left to gain.
Peak normalization is not loudness normalization
Peak normalization looks at one number: the highest absolute sample value in the file. Loudness normalization, the kind streaming platforms apply, measures perceived loudness over time using the LUFS scale defined in ITU-R BS.1770 and weights the measurement to match how hearing works.
The two can disagree wildly. A sparse acoustic recording with one loud transient will peak at full scale while sounding quiet throughout, and a dense mastered track can measure the same peak while sounding twice as loud. This tool performs peak normalization only, and says so plainly rather than implying it will match a track to a streaming target.
If your goal is consistent loudness across an album or a podcast feed, peak normalisation is a useful first pass but not the finished job. It gets every file into the same neighbourhood, and a loudness meter finishes the work.
Choosing a target level in dBFS
Full scale is 0 dBFS, the largest value a sample can hold. Normalising exactly to 0 leaves no headroom at all, which causes trouble later: lossy encoders reconstruct a waveform that can slightly exceed the original peaks, so an MP3 made from a file peaking at 0 dBFS can clip on playback even though the source did not.
Minus one dBFS is the common compromise for a finished file, and minus three is the usual target for anything that will be edited further, mixed with other material or encoded to a lossy format. Spoken word intended for further processing is often taken to minus six.
The panel shows the measured peak before and the calculated gain, so you can see whether the file needs a two decibel lift or a twenty decibel one. A very large required gain is worth noticing, because it also lifts the noise floor by the same amount.
How this audio normalizer works in your browser
Three steps run in the page. First, decodeAudioData converts your file into floating point sample arrays. Second, a scan across every channel finds the maximum absolute value, which is the peak. Third, an OfflineAudioContext renders the buffer through a gain node set to the ratio between your target and that peak, and the result is written to WAV.
Because the gain is applied during an offline render rather than during playback, the change is printed into the samples. The file you download is loud on its own terms, in any player, on any device, with no metadata tag required.
None of this touches a server. There is no upload request, no processing queue and no size limit beyond the memory available to your browser tab, which is what makes the tool usable on confidential interview recordings.
Normalising speech, music and field recordings
Speech benefits most. Interview and voice memo recordings are frequently captured twenty decibels below where they should be, and a single gain stage fixes that completely without touching the character of the voice.
Music needs more care. A track that already sits near full scale has nothing to gain, and a track with one stray transient will normalise to that transient rather than to the body of the music, leaving it quieter than expected. Trimming the stray peak first gives a far better result.
Field recordings sit in between. Lifting a quiet nature recording by thirty decibels also lifts the hiss by thirty decibels, so noise reduction belongs before normalisation rather than after it.
| Target | Headroom | Suits |
|---|---|---|
| 0 dBFS | None | Rarely correct, risks clipping after encoding |
| -1 dBFS | 1 dB | Finished files ready to publish |
| -3 dBFS | 3 dB | Material heading into further editing |
| -6 dBFS | 6 dB | Spoken word stems and multitrack sources |
How to normalize audio with this audio normalizer
- 1
Load the recording
Drop an MP3, WAV, OGG, FLAC, M4A or AAC file onto the panel. It is decoded in the page and never uploaded.
- 2
Read the measured peak
The tool scans every sample and reports the current peak in dBFS, which tells you how much headroom the file has.
- 3
Choose the target ceiling
Minus one dBFS for a finished file, minus three for anything that will be edited or encoded further.
- 4
Render and download
Press normalise, check the reported gain and the new peak, then download the WAV with the level printed in.
Related tools worth bookmarking
Sources and further reading
- MDN: OfflineAudioContextThe rendering context used to apply the calculated gain to every sample faster than real time.developer.mozilla.org
- ITU-R BS.1770 loudness recommendationThe standard that defines LUFS loudness measurement, the method peak normalisation does not use.itu.int
- W3C: Web Audio GainNodeSpecifies the gain node this tool uses to scale every sample by a single multiplier.w3.org
Frequently asked questions
Common questions about the audio normalizer.
No, provided the target stays at or below full scale. Multiplying every sample by the same factor is a mathematically clean operation. Distortion only appears if the resulting peak would exceed the maximum value a sample can hold, which the target ceiling prevents by construction.
No. Every sample is multiplied by the same number, so the ratio between the quietest and loudest moments is preserved exactly. If you want the quiet parts brought closer to the loud parts, you need compression rather than normalisation, which is a different operation entirely.
Because lossy encoders reconstruct a waveform that can overshoot the original sample peaks slightly. A file sitting exactly at full scale can therefore clip once it becomes an MP3 or an AAC. Leaving one to three decibels of headroom avoids that entirely.
No. Streaming platforms use LUFS loudness measurement, which weights frequency and averages over time. This tool performs peak normalization, which looks only at the single highest sample. The two are related but frequently disagree by several decibels on real material.
Yes. If the measured peak is already above your chosen target, the calculated gain is negative and the file is attenuated instead. That is useful for pulling a hot master back to leave room for further processing.