
Bring quiet and loud interview tracks to -16 LUFS with the Podcast target, read the gain summary, then merge them evenly.
You recorded three remote interviews for one episode. The first guest used a studio microphone and sounds loud, the second used a laptop and sits far back, and the third whispers. Cutting them together makes listeners ride the volume knob. The Audio Normalizer measures each file's loudness and applies the single gain that lands it on a standard target.
The plan is simple: normalize each guest's file on its own to the Podcast target of -16 LUFS, then join the files. With example readings of -26, -21 and -18 LUFS, the tool would apply +10, +5 and +2 dB respectively, and all three leave at the same perceived level.
The tool changes the whole file by one constant amount, up or down. It does not squash dynamics the way a compressor does, so a guest who mumbles in one sentence and shouts in the next still mumbles and shouts. What it fixes is the difference between files and between sessions. It uses a loudness measurement (EBU R128, in LUFS) that approximates how loud something sounds, rather than only how high its single tallest peak is.
That distinction matters. Peak normalization raises the loudest instant to a ceiling, so one cough can keep a whole quiet recording quiet. Loudness normalization looks at the whole file.
The tool offers five targets as buttons, with the figure shown beneath. Podcast / Spotify is -16 LUFS with a -1.5 dB peak ceiling. YouTube / streaming is -14 LUFS with -1 dB. Broadcast (EBU R128) is -23 LUFS. Audiobook (ACX) is about -20 LUFS with peaks held to -3 dB. Peak to -1 dB is the older method: it makes the file just louder without measuring loudness.
For this episode, choose Podcast / Spotify. If you were preparing a video soundtrack for streaming, YouTube / streaming at -14 would be the match. Always check the requirement of wherever the audio will be published, because platforms revise their guidelines.
Sometimes the tool can't reach the target. Say the whisperer has a few sharp peaks at -3 dB. Pushing the whole file up 10 dB would clip those peaks, so without help the tool stops at the safe limit and reports that the loud peaks prevent it going louder. The result is quieter than the others.
The checkbox labelled "Limit loud peaks so the loudness target can be reached" turns on a limiter. It lets the tool apply the full gain and catches the few peaks that would clip. Use it when a loudness target matters more than keeping every peak untouched, which is typical for speech. It is not used for the Peak target.
If the gain needed is more than 12 dB, the summary adds a note that the boost raises background noise too. That is your cue to listen for hiss, room echo and keyboard sounds in the quiet passages. Raising a -30 LUFS file by 14 dB lifts the noise floor by the same 14 dB. If it is too noisy, clean it with the Noise Remover before normalizing, or ask the guest to re-record closer to the microphone.
The summary for the Peak target reads differently: it states the loudest point before and after, such as "Loudest point -8.0 dB → -1.0 dB".
Trim first. A long silence or a rehearsal burst of laughter changes the measurement, since loudness is taken over the whole file. The Audio Trimmer cuts the heads and tails so the file you measure is the file you publish.
Normalize each source before merging, not the merged file, whenever the sources differ. A merged file gets one gain for everything, so the gap between guests remains. After merging, a final check on the combined episode is reasonable, and if you need a different output format afterwards, use the Audio Converter.
Normalize once at the end of your chain. Repeating it on already normalized files does little, but each re-encode of a lossy format can cost a little quality.
Normalizing to -14 for a podcast host that expects -16, or the reverse. Match the destination.
Expecting loudness normalization to fix inconsistent speech. A constant gain can't level a mumble against a shout.
Ignoring the 12 dB noise warning and publishing hiss.
Judging by waveform height. Compare by ear and by the LUFS figures in the summary, at the same playback volume.
Skipping the ACX difference. Audiobook uploads have their own rules, and the post on checking ACX audiobook levels covers that route.
The audio is processed in your browser, so the files are not uploaded to a server. Each normalization counts toward your 10 trial usages a day, so three guests means three runs. All the sound tools live in the audio tools category.
Are you normalizing a single voice, a multi-guest show or music for a video? Tell us which on the contact page.