LUFS explained: what −16 LUFS means for podcasts and social video

What LUFS measure

LUFS stands for loudness units relative to full scale; LKFS (loudness, K-weighted, relative to full scale) is the same measurement under the name used in North American broadcasting. Both come from ITU-R BS.1770, the standard behind modern loudness meters. One loudness unit (LU) equals one decibel, and readings are taken relative to digital full scale, so they are negative and a higher number means louder: −16 LUFS is louder than −23 LUFS.

What sets LUFS apart from a plain level meter is the weighting. The audio is first filtered roughly the way the head and ears hear it (the deep bass counts for less, the treble slightly more), then its energy is averaged over time. Two files with the same peak level can be several LU apart, while two files at the same LUFS sound about equally loud.

Momentary, short-term and integrated

Loudness meters show three readings, defined by the EBU alongside its R128 recommendation: momentary loudness over the last 0.4 s, short-term over the last 3 s, and integrated loudness over the whole programme. Targets refer to integrated loudness. It is gated: blocks quieter than −70 LUFS are ignored, then so is anything more than 10 LU below the average of what remains, so pauses and silences don’t drag the figure down.

True peak, and why −1 dBTP

A file’s samples are points on a curve. When a player rebuilds the curve for playback, it can swing higher between two samples than either of them: an inter-sample peak that a plain sample meter misses. True peak, written dBTP, estimates that real maximum by oversampling the signal, as BS.1770 describes.

Lossy encoders such as AAC, MP3 and Opus add small overshoots of their own when the sound is decoded. That is why distribution guidance usually asks for true peaks no higher than −1 dBTP: the decibel of headroom keeps the encoded file from clipping when it plays. US broadcasting asks for −2 dBTP, for the same reason.

Why loudness normalization matters

Streaming platforms and podcast apps measure each file and adjust its playback level toward a reference, so listeners don’t reach for the volume between two episodes or two videos. Uploading louder than that reference gains nothing: the platform turns the file down. Uploading much quieter has a cost: some platforms leave a quiet file quiet, so it sounds weak next to everything around it, while others turn it up, and the background noise rises with the voice.

That is why cleaning comes before normalizing. Raising a noisy recording to a normal level raises its noise by exactly the same amount, and a result that comes out too quiet is a frequent complaint about noise removers that stop at the cleaning.

Loudness targets compared

The figures differ because the listening differs. Broadcast keeps wide headroom for music and effects played in a living room; streaming and podcasts sit louder because people listen on phones, in transit, through earbuds. Short-video apps don’t publish an official target, and −16 LUFS holds up there too: loud enough for a phone speaker, and a platform that normalizes lower simply turns it down.

The mono nuance

Most voice recordings are mono: one channel. A loudness meter measures a mono file on that single channel, but players usually send it to both speakers or both earbuds, where the two add up: played on both channels, the same file measures about 3 LU louder than its reading. That is why some podcast guidance suggests about −19 LUFS for mono and −16 LUFS for stereo: two figures that sound alike in a pair of earbuds.

Noise Remover keeps −16 LUFS for a mono export, on purpose. It is a loud-enough target for phones and social video, close to the −14 LUFS reference of music streaming and YouTube, and any platform that normalizes lower turns it down at no cost to quality. If a podcast host or a client asks for −19 LUFS mono, lower the file by 3 dB in an audio editor before you upload it. More on preparing an episode on the podcasts page.

A stereo file is measured as stereo and reaches the same −16 LUFS; if you fold it to mono in the export options, it is measured again as mono, so the file you export is the one that is measured.

How Noise Remover normalizes

Noise Remover measures the final mix (the cleaned voice at the intensity you chose, over the trimmed range) with a BS.1770 meter built into the app, which reports both integrated loudness and true peak. It then applies one gain: the one that reaches −16 LUFS, unless it would push the true peak past −1 dBTP, in which case the ceiling wins. The preview already plays at that gain, so what you hear is what you export, and the original plays at about the same loudness, so the Original / Cleaned switch compares the noise, not the level.

There is no compressor and no limiter. A single gain changes the level without touching the dynamics: the voice keeps the natural rise and fall of speech, and nothing is squashed to reach a number.

Why a result can land below −16 LUFS

A file reaches −16 LUFS only if its true peaks sit no more than 15 dB above its integrated loudness, the distance between the target and the ceiling. Speech with sharp peaks, such as a plosive, a laugh, a word hit hard or a door closing, has a wider gap, so the gain stops at the ceiling and the export lands a few LU below the target. That is not a fault: a few LU is a small difference to a listener, and players that turn quiet files up close the gap.

If you need the full target (a podcast host that checks, a client’s specification), run the export through a compressor or a limiter in a desktop audio editor: lowering the peaks leaves room to raise the whole file. Don’t raise the gain past the ceiling without one, or the peaks will clip. The ceiling is measured before the final encoding (AAC, MP3 or Opus), which can add a fraction of a decibel; the headroom is there to absorb it.