Loudness normalization explained
Your recording is too quiet, or jumps between clips. LUFS is the number the pros use to fix that — here is what it means in plain English.
Peak level is not loudness
For years people judged audio by its peak level — the highest sample value, measured in dBFS. That number is easy to read but misleading, because it says nothing about how loud a recording actually feels. A file can peak near 0 dBFS yet sound quiet overall if most of the signal sits far below that peak.
Loudness measurement was invented to match human perception. LUFS — Loudness Units relative to Full Scale — weights frequencies the way our hearing does and averages over time. Two files with identical peaks can differ by many LUFS, and the one with the higher LUFS will sound louder to a listener.
The target numbers
Different platforms have converged on a few common targets:
- −16 LUFS — spoken-word podcasts. This is what most podcast platforms expect.
- −14 LUFS — general music streaming and social video.
- −23 LUFS — traditional broadcast television and radio.
These are integrated loudness targets: an average across the whole file. A true-peak ceiling, usually around −1.5 dBTP, is applied at the same time to prevent the peaks from clipping after the gain is raised.
What normalization actually does
Normalization measures the integrated loudness of a file and applies a single gain change to hit the target. It does not compress the dynamics, squash the peaks, or "make it punchy" — it simply moves the whole recording up or down to the right level while keeping a safety ceiling on the peaks.
This is different from turning the volume up in a player. Player volume changes at playback time and can push peaks into clipping; normalization changes the file itself, with a defined headroom. It is also different from compression, which reduces the gap between loud and quiet parts. If you need dynamics control, that is a separate process.
When it helps — and when it cannot
Normalization is ideal for a folder of clips recorded at different levels: it brings them all to one consistent loudness so a listener does not reach for the volume knob between tracks. That is the single most common real-world use.
It cannot rescue a recording that was captured badly. If someone spoke very quietly into a distant microphone, raising the overall gain also raises the noise floor. If the audio is distorted at the source, normalization will not undo the clipping. And it will not make a noisy recording cleaner — for that, run the noise reducer first, then normalize.
A practical workflow
For voice recordings, a sensible order is: clean up steady noise, even out the volume, then convert to your delivery format. The loudness normalizer on this site offers podcast, streaming and broadcast targets directly, so you do not need to remember the numbers. It runs locally in the browser with no upload.
If you are starting from an AMR voice note, remember the source is narrowband speech at a very low bitrate. Normalization will make it sit correctly alongside other audio, but it will not add frequency range that was never captured. For a text version as well, the transcription tool accepts the original file.
Loudness versus file size
A normalized file is not a smaller file. The process changes sample values, so the audio still compresses to roughly the same size in MP3 or M4A. If you want a smaller deliverable, pick a lower bitrate — but for speech, going below 96 kbps MP3 or AAC starts to add audible artefacts that are far more noticeable than a few LUFS of level difference.
It is also worth knowing that many platforms normalize playback themselves. Upload a very quiet file and Spotify or YouTube will turn it up to their own target anyway. Normalizing before you upload still helps: it prevents your track from sounding out of place in playlists that have gain matching switched off, and it gives you a predictable result everywhere else.
Checking the result
After normalizing, listen at a moderate volume alongside a reference track you know sits at the right level. Your ears are the final check — a number on a meter cannot tell you that a recording sounds unnatural. If it now feels too loud or too quiet relative to your other content, adjust the target rather than applying a second gain change on top.