"Normalize" means applying a gain change so that some measurement of the audio reaches a target value. The measurement can be the highest peak or the average loudness, and that choice makes a big difference to the result.
Decibels in one minute
Digital audio is measured in dBFS — decibels relative to full scale. 0 dBFS is the largest value a file can store; everything else is negative. A few numbers worth remembering:
- +6 dB doubles the amplitude; −6 dB halves it.
- +10 dB is heard as roughly "twice as loud".
- A typical phone voice memo peaks around −12 to −6 dBFS; a commercial pop master peaks near −1 to 0 dBFS.
Anything pushed above 0 dBFS is clipped: the tops of the waveform are cut flat, which sounds harsh and crackly.
Peak normalization
Peak normalization finds the single loudest sample and calculates the gain that would bring it exactly to the target. That gain is then applied to every sample equally.
Because the loudest point lands at the target, peak normalization can never clip. And because one gain is applied everywhere, the relationship between quiet and loud passages is unchanged — it is exactly like turning a volume knob to the highest position that doesn't distort.
The weakness is that one stray spike controls everything. A cough, a microphone bump or a single snare hit 12 dB above the rest of the recording means the rest can only come up by the same small amount. The fix is simple: trim the spike out of the selection, or normalize just the part you need. In our normalize tool, normalization is measured on the current selection only, so a loud moment outside it doesn't count.
Why −1 dBFS and not 0?
MP3 and AAC encoders don't store samples; they store a description of the sound, and the decoder rebuilds the waveform. The rebuilt peaks can land slightly higher than the original — so-called inter-sample or true peaks. A file normalized to exactly 0 dBFS may therefore clip after encoding or on cheaper phone DACs. One decibel of headroom is a common, safe margin, and it is the target this site uses.
Loudness normalization (LUFS)
Peaks don't tell you how loud something sounds. A sharp drum hit and a sustained synth pad can have the same peak but very different perceived loudness. Loudness normalization measures average perceived loudness in LUFS (loudness units relative to full scale), a standard based on how human hearing weights different frequencies.
Streaming services use it to keep tracks at a consistent volume. Spotify and YouTube target around −14 LUFS, and Apple Music around −16 LUFS. Podcasts are commonly delivered at −16 LUFS for stereo.
| Peak normalization | Loudness normalization | |
|---|---|---|
| Measures | Highest sample | Average perceived loudness |
| Can clip? | Never | Yes, if it raises the level a lot, unless a limiter is used |
| Consistent loudness between files? | Not really | Yes |
| Best for | Ringtones, clips, one-off recordings | Playlists, podcasts, streaming delivery |
For a ringtone or a single voice clip, peak normalization is the right tool: you want it as loud as possible without distortion. Loudness normalization matters when many files need to sit together at the same level.
Normalization vs compression
Normalization never changes the dynamic range — the difference between the quiet and loud parts. If a recording has a whispered sentence followed by a shout, the whisper is still much quieter after normalizing. A compressor is the tool that reduces that difference, by turning down loud passages so the overall level can be raised further. Commercial music sounds loud largely because of compression and limiting, not normalization.
Using the volume slider instead
If you untick Normalize, the Volume slider applies a fixed gain from −12 to +12 dB. This is useful when you want a clip quieter, or when you want it louder than normalization allows and are willing to clip a rare spike. The tool shows a warning when the result will clip, so you can decide before exporting.
Practical recipes
- Quiet voice memo: trim the ends, tick Mono and Normalize, export MP3 at 128 kbps.
- Ringtone: select the chorus, Normalize, add a 1–2 s fade out. See iPhone or Android ringtone guides.
- Clip that is still too quiet after normalizing: find the spike (the tallest line in the waveform), move the selection to exclude it, or raise the volume a few dB and accept slight clipping on that spike only.