Audio Loudness Standards and Why Your Video Sounds Quiet

Your video sounds quiet next to everything else in the feed, so you turn it up, add a limiter, and push until the meters are hitting the top. It still sounds quiet — and now the voice has lost its shape.

The problem is that the meter you were watching is not measuring what listeners hear. Peak level tells you the height of the tallest instant in the file. Loudness is a perceptual quantity averaged over time, and the two are only loosely related. This guide covers what the loudness measurements actually mean, why platform normalisation makes the loudness war pointless, and how to set levels so a mix arrives sounding the way you left it.

Peak, RMS and LUFS

Integrated LUFS is the one to work to. It measures the whole piece, weighted to approximate the ear's sensitivity across frequencies, and it is the closest thing available to a number describing how loud something sounds.

The gap between peak and loudness is why a sparse voice recording can peak at maximum and still sound thin: the peaks are brief and most of the file sits far below them. Conversely a dense, heavily compressed mix with lower peaks can sound relentlessly loud, because its average energy is high.

Normalisation Changed the Rules

Most large video and audio platforms measure the integrated loudness of what you upload and adjust it toward their own target. The specific targets differ between services and are revised from time to time, which is why they belong on a delivery sheet rather than in your memory — the split between durable settings and policy ones is covered in export settings for each major platform.

The consequence is what matters here. If you deliver substantially louder than the target, the platform turns you down. You do not get to be louder than everyone else; you only get to arrive with whatever damage the loudness-chasing did to your dynamics, at the same playback level as a mix that was left alone.

That inverts the old advice. Rather than mastering as hot as possible, aim near the target you are delivering to, keep some dynamic range, and let normalisation do nothing. A mix that needs no adjustment is a mix that reaches the listener intact.

Set Dialogue First

For anything speech-driven, the useful working method is to treat dialogue as the reference and build everything else around it.

Get the voice sitting consistently first, using compression to even out the performance rather than to make it loud. The goal at this stage is that no line is noticeably quieter than another — that is what stops viewers reaching for the volume control.

Then bring music in underneath. The common failure is setting music by ear during an instrumental passage, where nothing is competing with it, and discovering it buries the voice everywhere else. Set music level while the voice is speaking, not during a gap.

Effects sit last, and they are the element most often left too loud, because they were auditioned in isolation.

Where music and voice genuinely overlap for long stretches, level alone will not solve it and the fix is spectral — carving space in the music where the voice lives. That is a separate technique covered in mixing voiceover with music.

Leave Headroom

True peak matters because of what happens after you deliver. Lossy encoding reconstructs the waveform slightly differently, and a file that just touches zero can overshoot it after re-encoding, producing distortion that was not in your master.

The standard precaution is to set the limiter's ceiling a decibel or so below zero rather than at it. That costs nothing audible and removes an entire class of problem that is otherwise only discoverable after publishing.

Why Compression Stopped Buying Loudness

The habit of mastering as loud as possible came from an era where whatever you delivered was what played. Louder genuinely was louder, and everyone pushed, which is how a decade of increasingly flat mixes happened.

Normalisation removed the incentive and left the habit. Once a platform measures your integrated loudness and adjusts it toward a target, the only thing heavy limiting achieves is arriving at the same playback level with less dynamic range than the person who did not bother.

It is worth being concrete about what is lost. Compression reduces the difference between the loudest and quietest parts. Applied gently that is useful — it evens out a performance and keeps quiet phrases audible. Applied hard it removes the emphasis a speaker put into their delivery, so every sentence lands with the same weight. Listeners do not identify this as compression; they experience it as the speaker sounding flat, or as the video being tiring to watch for more than a few minutes.

There is one place the old logic still applies: destinations that do not normalise. A file played directly from a website, a download, or an installation has no adjustment applied, so the level you deliver is the level heard. Even there, matching the loudness of comparable material is a better goal than being louder than it.

Measuring Rather Than Guessing

Loudness is one of the few things in production that is not a matter of taste, and it is worth using the meter rather than your ears — for the same reason colour work uses scopes. Your perception of loudness changes with how long you have been listening, how loud the room is, and what you heard immediately before.

Three readings tell you nearly everything. Integrated loudness across the whole piece is the figure to compare against a target. Short-term loudness shows you which passages sit high or low against the average, which is how you find the section where the music took over. And loudness range tells you whether the piece has any dynamics left at all — a very low range on speech content is a sign of over-compression rather than of consistency.

Take the readings on the finished mix, in one pass, with nothing soloed. A measurement taken over a section is not the number the platform will calculate.

Where the Chain Sits

Loudness work is the end of the audio process, not a substitute for it. A recording with a high noise floor, harsh sibilance or uneven room tone will be a louder version of those problems after this pass — the sequence of repairs that has to happen first is in the order to clean up bad audio.

It is also the point where a mix is easiest to ruin. Everything up to here can be adjusted; a limiter applied hard at the end changes the material irreversibly in the render. Keep the limiting gentle, and if you find you need a lot of it, the problem is a balance issue further back.

One practical note on delivery. Export your audio at the mix level rather than pre-normalising for a specific platform, and keep the master. Different destinations have different targets, and a file already squashed to meet one of them cannot be un-squashed for another.

Next Steps

Measure the integrated loudness of your last three published videos with any loudness meter — most editors include one, and standalone tools are freely available. If the numbers vary widely between them, that inconsistency is what viewers experience as some of your videos being quiet.

Then set a single target for your channel, aimed at the platform you publish to most, and check every export against it before delivery. Consistency across your own library matters more to a returning viewer than hitting any particular figure. The rest of the delivery material sits under formats, specs and delivery.