What is loudness in video audio?
Loudness is how loud audio actually feels, measured in LUFS. It differs from peak level — a track can peak at 0dB and still feel quiet. YouTube and most platforms normalize playback to around −14 LUFS, so mastering speech near that target with true peaks under −1dB is the practical goal.
Why do voices pump in AI-edited video?
Because loudness got automated with the wrong tool. Dynamic normalization (ffmpeg's loudnorm and its cousins) continuously adjusts gain toward the target. Music masks the movement; speech doesn't. Every pause lets the gain climb, every sentence knocks it down — the audible result is a voice that breathes in rhythm with its own sentences.
The boring fix wins: measure once, apply a static gain, catch the peaks with a limiter. One number decided up front, nothing moving during playback. It's less clever than an adaptive loop, and it's what a mastering engineer would do to a voice track.
Loudness targets that matter in practice
| Context | Target | Why |
|---|---|---|
| YouTube / TikTok / Reels | ≈ −14 LUFS int. | platform playback normalization |
| Podcast | −16 LUFS stereo / −19 mono | listening in noise, earbuds |
| True peak, everywhere | ≤ −1 dBTP | lossy encoders overshoot above it |
Read the transcript
What is loudness in video audio? Loudness is how loud audio actually feels, measured in LUFS. It differs from peak level — a track can peak at 0dB and still feel quiet. YouTube and most platforms normalize playback to around −14 LUFS, so mastering speech near that target with true peaks under −1dB is the practical goal.
Frequently asked questions
What LUFS should I target for YouTube and TikTok?
Around −14 LUFS integrated for both, with true peaks at or below −1dBFS. Louder masters get turned down by the platform; quieter ones make viewers reach for the volume control.
What is pumping, and what causes it?
Pumping is the audible breathing of a level that rises in pauses and ducks on syllables. Dynamic loudness normalizers chase the target in real time, and on speech — all transients and silence — the chase itself becomes audible.
How does Never level speech without pumping?
A static chain: a fixed gain stage into a hard limiter, with the limiter's lookahead latency compensated so audio stays frame-aligned with video. No decision varies over time, so there is nothing to hear moving.