In this guide
Peak normalization sets the highest sample level with one gain adjustment. Loudness normalization uses a loudness measurement. Choose by the problem you hear, keep stereo balance deliberate, and measure the final export when a delivery specification matters.
If a recording is quiet throughout, peak normalization can raise its level with one gain adjustment. If you need several recordings to have consistent perceived loudness, use a loudness measurement and an intentional loudness-normalization workflow. These operations answer different questions, so start by listening to the problem rather than choosing the largest number.
The free audio normalizer provides sample-peak normalization for short mono or stereo MP3/WAV files. It reports the decoded peak, lets you choose a target in dBFS, applies the same gain to both stereo channels, and exports a WAV locally. It does not measure LUFS or true peak.
What peak normalization does
Peak normalization finds the largest absolute sample value and calculates a constant multiplier to bring it to the target. Every sample receives that multiplier. The timing, channel count, and relationship between quiet and loud passages stay the same before export quantization.
For example, a peak of 0.25 relative to full scale is about −12.04 dBFS. A target of −1 dBFS corresponds to a peak of about 0.89125. Dividing the target amplitude by the measured peak gives a gain of about 3.565, or +11.04 dB. The downloaded 16-bit WAV can differ slightly because its sample values are rounded.
| Original sample peak | Target sample peak | Required gain |
|---|---|---|
| −18 dBFS | −3 dBFS | +15 dB: raise the complete clip |
| −6 dBFS | −1 dBFS | +5 dB: raise the complete clip |
| −1 dBFS | −6 dBFS | −5 dB: lower the complete clip |
These are arithmetic examples, not measurements of recordings or recommended delivery standards. A target should fit the edit you are making. Increasing the level can make an already clear, quiet clip easier to place in a project; decreasing it can leave more sample headroom for a later edit.
What loudness normalization adds
A highest sample does not describe the whole listening experience. A short impact and a dense, sustained passage can have the same peak. The quiet recording around that impact may still seem much softer. Matching both files to the same sample peak leaves their different patterns intact.
Audacity’s Loudness Normalization reference distinguishes perceived-loudness measurement in LUFS from RMS measurement. Use an appropriate loudness meter when your task is to match perceived loudness or satisfy a stated loudness requirement. Sample-peak dBFS is not a substitute for a LUFS reading.
Even a loudness target does not automatically resolve every delivery issue. The final file’s peak and changing levels still matter. FFmpeg’s loudnorm documentation, for example, describes separate integrated-loudness, loudness-range, and maximum-true-peak controls. A single gain setting is not the same process.
Read the actual delivery specification instead of applying a universal “podcast” or “streaming” preset to every project. This free tool offers a sample-peak target only. For a LUFS requirement, finish in an editor or tool that measures and processes loudness, then verify its final export.
When a peak adjustment is a useful starting point
| Problem | Starting workflow | Why |
|---|---|---|
| The complete clip is quiet, with peaks well below the desired level | Test a peak-normalized copy | One gain can raise the whole recording while retaining its internal balance |
| One impact dominates a quiet interview | Inspect the impact and edit it deliberately before remeasuring | The highest sample determines a peak normalizer’s gain |
| Two speakers alternate between quiet and loud sentences | Use separate track gain where available, or careful automation/dynamics processing | One gain leaves their relative difference intact |
| Several clips need consistent perceived loudness | Use loudness measurements and a chosen loudness target | Their peak values alone do not describe their overall level |
| Noise covers the voice | Evaluate noise reduction or source isolation | Increasing gain changes the voice and noise together |
Keep any edits to impacts or individual words separate from peak normalization. A compressor, limiter, or gain envelope can change different parts by different amounts. Their settings and audible tradeoffs need their own listening checks.
Preserve stereo balance deliberately
This normalizer measures the highest sample across both channels and applies one shared gain. Suppose the left peak is 0.25 and the right peak is 0.125. Bringing the left peak to about 0.89125 brings the right to about 0.44563. The right remains about 6.02 dB below the left in sample peak.
That is useful when the existing stereo balance is intentional. Audacity’s Normalize documentation explains the distinction between linked and independent stereo-channel processing. This tool always links channels and does not remove DC offset.
If an interview has a different microphone on each side, decide whether the channels should be edited separately first. The channel splitter can export existing left and right recordings as separate mono WAVs. Channel extraction still does not remove room noise or spill already captured on each microphone.
Gain does not improve the voice-to-noise relationship
A constant multiplier applies to every sound. If you raise a voice by 10 dB, you also raise the hiss and traffic mixed with it by 10 dB. A recording may become easier to hear, but the relative level of that noise has not improved.
Listen before normalization to decide whether the clip is simply quiet or whether a competing source is the real obstacle. The audio cleanup chooser helps distinguish a steady-noise treatment from source isolation, a gate, or an edit. For an overlapping source, evaluate the voice isolation workflow on a representative original excerpt.
After cleanup, compare the processed result with the original at a similar listening volume so the louder version does not dominate your judgment. Check consonants, soft words, breaths, and overlapping sections. Choosing the same peak for both copies is not a complete loudness match; a loud impact can still skew the comparison.
Sample headroom is different from a true-peak check
A −1 dBFS sample target leaves room below the full-scale stored-sample limit. It does not establish the highest reconstructed peak between those samples. Later mixing, effects, and lossy encoding can also change the final result.
For a delivery with a true-peak limit, use a true-peak meter on the final exported file. This tool does not oversample to estimate true peak or use a limiter. Do not turn its default target into a claim that every playback system or later encoding is protected from overload.
Lowering a damaged clip is also not a clipping repair. Flat or distorted portions already in the source remain after gain adjustment. An alternate intact take may be a better starting point than repeated processing.
Normalize and check a short file locally
- Open the peak normalizer and choose a supported MP3 or WAV up to 50 MiB and five minutes.
- Play the original at a comfortable level. Read the measured decoded sample peak and check whether the issue is overall gain.
- Enter a sample-peak target from −24 to −0.1 dBFS. Read the required gain; a large boost also raises low-level noise.
- Prepare the normalized WAV. Alternate between the original and result; the players pause each other to help avoid doubled playback.
- Download the result and listen in the intended edit. For a specified delivery, measure the final export with appropriate loudness and true-peak tools.
The output is 48 kHz, 16-bit PCM WAV, retaining the complete decoded duration and mono/stereo channel count. Browser decoding can resample the input, as documented in MDN’s decodeAudioData reference. This is a new audio export, with no MP3 encoding or metadata copy.
A compressed input can fit while its uncompressed WAV is too large. The output must also fit within 50 MiB. Use the size calculator to plan a shorter excerpt, or the trimmer to cut one. The WAV inspector can check the downloaded file’s stored format and size. Keep the original throughout.
Common questions
Why did my normalized clip remain quiet?
One loud sample or impact can determine the gain for the entire clip. Inspect that event and the changing level of the wanted audio. A higher sample target alone may leave the passages you care about relatively quiet.
Can I normalize without uploading audio?
Yes. The free normalizer decodes, measures, adjusts, previews, and exports in this browser. No audio is sent to Saturalabs, and no account or download credits are required for this utility.
Should I normalize every file before AI isolation?
Do not assume a gain change improves separation. If your supported original is ready, test it first and judge the audible result. The preparation guide explains how to choose a useful excerpt and preserve its timeline position.
Does equal peak mean equal volume?
Equal sample peak means the maximum stored sample reaches the same level. It does not ensure that different recordings have equal perceived loudness. Use a suitable loudness measurement for that comparison.