The short answer
Loudness normalization is something platforms do to your audio after you hand it over. Before a service plays your track next to everyone else's, it measures how loud your whole program actually is, as one number: its integrated LUFS. Then it turns the playback up or down so your program lands at the same loudness as everything around it. Loud masters get turned down. Quiet ones often get turned up. The listener never has to ride the volume knob between tracks.
The key word is playback. Normalization does not touch your file or re-master anything. It computes a single gain offset and applies it while your audio plays, the same way you would nudge a fader. Your dynamics, your peaks, and your mix are all untouched; only the overall level the listener hears is matched to a target.
How playback normalization works
Every service that normalizes does the same three things. First it measures your program's integrated loudness over its whole length, using the LUFS standard (ITU-R BS.1770), which weights the audio roughly the way your ears do and averages it across the program with the silent parts gated out. That gives one number, say -9 LUFS for a hot master. Second, it compares that number against its target, for example -14 LUFS. Third, it applies the difference as a playback gain offset.
playback_gain_dB = target_LUFS - measured_LUFSBecause 1 LU equals 1 dB, the offset is just the distance from your measured loudness to the target.
A program measured at -9 LUFS against a -14 target gets -14 minus -9, which is -5 dB: it plays back 5 dB quieter than it was mastered. A quiet program at -20 LUFS gets +6 dB and plays back louder. A program already sitting at -14 gets 0 dB and is left exactly as it is.
Nothing here re-encodes or squashes your audio. The offset is a level change applied at play time, so a program mastered with wide dynamics keeps every bit of them; it just sits at the shared loudness alongside everyone else.
Why this ended the loudness war
For two decades, mastering got louder and louder. The logic was crude but real: on a playback system at a fixed volume, the louder track jumps out, feels bigger, and wins the quick comparison. So engineers pushed levels up with heavy limiting and clipping until masters were brickwall flat, their dynamics crushed out to squeeze the average up a few more dB.
Normalization quietly ended that arms race. If the platform is going to measure your integrated loudness and turn a -9 LUFS master down to -14 anyway, the loudness you fought for simply evaporates on playback. What you are left with is the damage: a squashed, fatiguing master that now plays at the same loudness as a track with its dynamics intact, and sounds worse doing it. Turning something down does not restore the peaks you crushed to get it loud. So the winning move flipped. Master for the target, keep your dynamics, and let the quiet to loud contrast breathe.
Overshooting and undershooting the target
Once you know the platform will move your level to its target, the only questions are which direction it moves you, and what that costs.
- Master louder than the target and the platform turns you down. You gain no loudness, because everyone lands in the same place, and you keep whatever dynamic damage you did chasing the level. Worse, a hot master jammed against the ceiling risks true-peak overshoots: inter-sample peaks that can clip after the platform's codec re-encodes your audio. That is why masters usually leave about a dB of headroom, the same reason a limiter ceiling sits below zero.
- Master quieter than the target and behavior splits. Some services turn you up to meet the target, which can lift the noise floor and any hiss along with it. Others only ever turn programs down and leave quiet ones quiet, so undershooting can leave you audibly softer than the tracks around you. Aiming near the common target is the safe bet either way.
The targets are not all the same
Different destinations settled on different numbers, so "the target" depends on where your audio is going. Music streaming services (Spotify, YouTube, Apple Music) normalize to roughly -14 LUFS. Spoken-word and podcast targets sit a little lower, around -16 LUFS, though some platforms use -14. Broadcast is quieter still: the EBU R128 standard targets -23 LUFS and ATSC A/85 targets -24 LKFS (LKFS and LUFS are the same unit, just different names). The full picture, and how to pick one, is in LUFS targets for streaming and podcasts.
Loudness normalization in patchd
Here is the honest part: patchd does not normalize your audio, and it has no LUFS or loudness meter. Normalization is a program-loudness decision made downstream, by the streaming service, podcast host, or platform that plays your finished audio to an audience. patchd sits upstream of all of that, in real time, while you are still talking or playing.
What patchd gives you is clean, controlled levels going in. Each channel meter shows peak level in dBFS, a live per-channel readout from -60 up to 0 (silence reads as -infinity) fused with a VU meter in the fader, so you can see exactly how hot each source runs and set your gain with headroom to spare. The free limiter node holds a hard ceiling, default -1.0 dB, so a stray peak never clips on the way to whatever records or streams you. Because it runs in a live monitoring path it uses zero lookahead and adds zero latency, which means it catches sample peaks rather than true (inter-sample) peaks; the platform's own measurement takes it from there.
In other words: you deliver a tidy, well-staged signal at a sensible level, and the platform's normalization carries it the rest of the way to its target. Getting your gain staging right upstream is what makes that hand-off clean.
The takeaway
Loudness normalization is playback gain matching: the platform measures your program's integrated LUFS, subtracts it from its target, and applies that offset so everything plays at one loudness. It made the loudness war pointless, because a brickwall master just gets turned back down and keeps its crushed dynamics for nothing. Master for the target with your dynamics intact, leave about a dB of headroom for peaks, and let the platform match the levels.
patchd is a pre-launch Windows audio mixer that keeps your levels clean and honest before they ever reach a platform: per-channel dBFS metering, a free limiter on every channel, and a real-time engine that adds only single-digit milliseconds at a small ASIO buffer. If that is the front end you want feeding your stream or recording, join the waitlist and we will tell you when it ships. patchd is in development now; the Studio tier is $39.99 per year, and the real-time engine is identical on both tiers.