The math of LUFS - getting your audio loud enough without clipping

If you've ever uploaded a video that was perfectly fine in your editor and then sounded weirdly quiet on YouTube, that's not a bug - that's loudness normalization. Every major platform measures your audio against a reference level called LUFS, and if you're below it, the platform turns you down. Above it, it turns you down harder.

This is the working approach for getting the math right before you upload.

What LUFS is, in one paragraph

LUFS - Loudness Units relative to Full Scale - is a perceptual measure of how loud something feels, averaged over time. It's different from peak level (the single loudest sample) and different from RMS (a simpler averaging method). LUFS is what platforms actually measure to decide whether to turn your video up or down before serving it to the listener.

The whole game is to deliver audio at the platform's target LUFS, with peaks below the platform's clipping threshold. Hit those two numbers and your video plays at the same perceived loudness as everything around it.

The platform targets

The numbers you need to know:

• YouTube - −14 LUFS integrated, −1 dBTP True Peak ceiling
• Spotify - −14 LUFS integrated, −1 dBTP
• Instagram / Reels - −14 LUFS integrated, −1 dBTP
• TikTok - roughly −14 LUFS integrated, but less strictly enforced
• Apple Music - −16 LUFS integrated (the only major outlier on the quiet side)
• Podcast - −16 LUFS integrated, −1 dBTP

For video, −14 LUFS is the modern standard. If you master to −14 and peak at −1, your audio plays at the platform's intended loudness without normalization knocking it down.

What happens when you miss

Master at −18 LUFS and upload to YouTube. The platform decides you're 4 LUFS quiet relative to the rest of the catalog. It doesn't turn you up - that would risk clipping. It just leaves you quiet, and your viewer cranks their volume. Now your quieter moments are also louder than the surrounding catalog's quieter moments, which means your dynamic range is fine but your average is lower than the next video. Statistically, viewers swipe.

Master at −10 LUFS and upload. The platform decides you're 4 LUFS too loud. It turns you down by 4 LUFS, and now everything - voiceover, music, sound design - drops by the same 4 LUFS. Your carefully balanced mix is exactly the same shape, just quieter. You gained nothing by being loud.

The lesson: the only loudness that matters is the one the platform delivers. Hit −14 and stop pushing.

Peak versus true-peak

dBFS (full-scale peak) measures the loudest sample in the digital signal. dBTP (True Peak) estimates the loudest inter-sample peak - the moment between digital samples when the analog waveform might exceed 0 dB.

Why it matters: most encoders (AAC, MP3, Opus) introduce small inter-sample peaks during conversion. A file that peaks at 0 dBFS in your DAW can peak at +0.6 dBTP after encoding. That overshoot clips on consumer playback, and clipping is the single most audible flaw.

Master with a ceiling of −1 dBTP, not 0 dBFS. The 1 dB of headroom absorbs the encoder's overshoot.

A simple master chain

For a typical explainer with voiceover + music bed + occasional sound design:

1. Voiceover bus
   - High-pass at 80 Hz
   - De-esser at 6-7 kHz (3-5 dB reduction)
   - Compressor: 3:1, 8ms attack, 80ms release
   - Trim to -16 LUFS short-term

2. Music bus
   - Ducking sidechain from voiceover (4-6 dB reduction)
   - Trim to -22 LUFS short-term when voice is present

3. Master bus
   - Bus compressor: 2:1, slow attack, gentle (1-2 dB)
   - Limiter: ceiling -1 dBTP, release 50ms
   - Target integrated: -14 LUFS

The Ducking step is the one most amateur mixes skip. Without it, the music fights the voiceover and the viewer feels they have to lean in to understand the narration. With it, the music gets out of the way during dialogue and comes back during transitions.

Measuring it

You need a loudness meter that reports integrated LUFS - not just peak. Most modern DAWs have one built in (Logic's MultiMeter, Reaper's ReaLoud, Premiere's Loudness Radar). For a standalone option, MeterPlugs's Loudness Penalty plug-in is free and accurate.

Run the entire video through the meter from end to end. The integrated LUFS reading at the end is the one the platform will see. Short-term and momentary readings are useful for finding individual loud or quiet moments, but the upload decision is integrated.

The dynamic-range trap

A common mistake: pushing the integrated LUFS up by squashing dynamic range. The voiceover gets loud, the music gets loud, the sound design gets loud, and the whole thing reads as flat. The platform serves your flattened mix at exactly the same loudness as everyone else's, except yours has no shape.

A useful counter-metric: loudness range (LRA). For dialogue-driven explainers, an LRA of 6–9 LU is healthy. Below 5 means you've squashed it. Above 12 means you have moments the viewer will struggle to hear on a phone in a noisy room. Aim for 7.

Mono compatibility

Phones play through mono speakers. Mono compatibility is the property that your mix doesn't lose anything important when it collapses to mono.

The two pitfalls:

• Wide stereo music - a track with hard-panned high-hats can cancel to silence in mono. Test by listening with your DAW's mono button engaged.
• Reverb-only stereo width on dialogue - sounds spacious in stereo, sounds hollow in mono. Apply reverb to a separate bus and pan it more conservatively.

If your mix sounds different in mono than in stereo, the platform - and most of your viewers - will hear the mono version.

Quick checklist before export

• [ ] Integrated LUFS within ±0.5 of -14.0
• [ ] True peak under -1.0 dBTP
• [ ] Loudness range between 6 and 9 LU
• [ ] Mono playback retains all elements
• [ ] No clipping in the limiter (zero overshoots)

If all five are green, the upload will play at platform loudness without normalization adjustments.

When -14 isn't the answer

Three exceptions where the standard target doesn't hold:

• ASMR or whisper-driven content - master quieter (−18 to −20 LUFS) and trust the viewer to lean in.
• Music videos where the music is the point - match the loudness of the album master, even if it's louder than −14. The platform will normalize, but the relative dynamics matter more than the absolute loudness.
• Podcasts re-cut as video - keep the podcast master (−16 LUFS) rather than pushing for video standards. The audience is listening, not watching.

Bring it to VideoCue

The platform's render pipeline targets −14 LUFS integrated with a −1 dBTP ceiling by default. If you want to override for one of the cases above, the LUFS entry covers the measurement details, True Peak covers the peak-ceiling logic, and Ducking covers the sidechain trick that keeps music out of the voiceover's way.