Compression vs Normalization for Spoken Audio
Normalization changes the overall level. Compression changes the difference between loud and quiet moments. They solve different problems, and you may need both.
By the fix.fm team · Published · Updated
First ask whether the file is quiet or uneven
If every sentence needs the same lift, a uniform level change may be enough. If one sentence is barely audible and the next is very loud, raising the whole recording keeps that difference. The loud sentence reaches the ceiling first. Adjusting a separate speaker track or an isolated loud event may be more natural than putting a strong compressor across the whole conversation.
| Process | What it changes | What it does not solve by itself |
|---|---|---|
| Peak normalization | Uniform gain chosen from the highest peak | A quiet average or uneven sentences |
| Loudness normalization | Uniform gain chosen from a loudness target | Local balance when the target requires too much peak gain |
| Compression | Time-varying gain that reduces level differences | Missing speech, room noise or incorrect routing |
| Limiting | Strong dynamic peak control | A good balance between every speaker |
Those normalization descriptions refer to a basic gain-only process. A product called a normalizer may also compress or limit, so check what that particular tool does. Audacity documents peak normalization and loudness normalization as different level choices.
Work through two moments in the same recording
Imagine a quiet moment at -30 dBFS and a loud moment at -10 dBFS. Adding 5 dB brings them to -25 and -5: the gap remains 20 dB. A simple compressor with a -20 dBFS threshold and a 2:1 ratio lets each 2 dB above the threshold become 1 dB above it. The loud moment becomes -15, while the quiet moment stays -30. Add 5 dB afterward and the pair is -25 and -10, with a 15 dB gap.
| Change | Quiet moment | Loud moment | Gap |
|---|---|---|---|
| Original | -30 dBFS | -10 dBFS | 20 dB |
| Uniform gain, +5 dB | -25 dBFS | -5 dBFS | 20 dB |
| 2:1 compression, then +5 dB | -25 dBFS | -10 dBFS | 15 dB |
This arithmetic isolates the difference; it is not a compressor preset for your voice. Actual compressors choose a detector, smooth gain over time, and may use a soft knee. Attack controls how quickly gain reduction responds; release controls how it recovers. The Audacity compressor guide explains threshold, ratio and makeup gain. Processing can lower the output first, so compare with and without it at similar listening levels.
A loudness target still needs peak headroom
Suppose a finished file reads -23 LUFS and -3 dBTP. A gain-only change toward -16 LUFS would require approximately +7 dB and push that peak toward +4 dBTP. A ceiling of -1 dBTP allows only about +2 dB of uniform gain. That can get you roughly -21 LUFS, not -16.
You can choose a quieter delivery target where permitted, reduce the unusually loud event, or use dynamic control before the final measurement. Do not force the target by hard-clipping the samples. LUFS vs dBFS vs dBTP explains why the programme average and peak ceiling are separate readings.
Measure the actual delivery file afterward. Lossy encoding can alter peaks, and an integrated reading does not show every speaker's local level. A file can meet a programme target and still contain an answer that is difficult to hear.
Choose the tool by the change you want
For a custom whole-file loudness target with uniform gain, use the free volume normalizer on tonal.fm, our sister site. It respects its peak ceiling and reports when simple gain cannot reach the target. That is useful when you want to preserve the recording's existing dynamics.
The quiet-audio fixer works differently: it raises quiet speech using a gentle compressor and a lookahead limiter, toward our speech target. It does not independently mix remote speakers. If only one participant is quiet, work on that person's separate track before the final mix whenever possible.
The LUFS meter measures without changing the audio. Use it before deciding and after the edit. Compression is also unrelated to MP3 file compression: one changes dynamics, while the other discards information to make a smaller encoded file. WAV vs MP3 covers that separate decision.
Listen for the cost of making it more even
Too much dynamic control can bring breaths and room noise forward, flatten emphasis, or make the background rise and fall unnaturally. Start with a small change and listen to a quiet phrase, a loud phrase and a transition between them. If the problem is a single bump or shout, an individual edit may protect the surrounding speech better.
Keep the source, finish the content edit, and check the loudness suited to your destination. The podcast workflow places that decision after editing and repair, and the printable checklist helps you repeat the final checks.