Voice-Over Loudness Standards: LUFS Targets by Platform

If a client asks you to deliver at −23 LUFS and you have been mastering podcasts at −16, you are not doing anything wrong — you are working to a different standard. Voice-over sits across broadcast, film, streaming and podcasting, and each of those worlds settled on its own loudness target years ago.

This page collects the numbers in one place, with the source for each. Bookmark it, or steal the table.

LUFS, LKFS and dBTP — what the units measure

LUFS stands for Loudness Units relative to Full Scale. It measures perceived loudness using a frequency weighting that approximates how human hearing works, which is why it correlates with how loud something actually sounds rather than how tall the waveform is.

LKFS — Loudness, K-weighted, relative to Full Scale — is the same measurement under a different name. Broadcast and film documentation uses LKFS; music and podcast documentation tends to use LUFS. One LU equals one dB. If a spec sheet mixes the two, it is not describing two different things.

dBTP is true peak, measured with inter-sample peak detection. It is not the same as sample peak. A file that reads −0.1 dBFS on a sample-peak meter can easily exceed 0 dBTP once it is converted to analogue or encoded to a lossy format, which is where the distortion appears.

Three other terms show up in specs. Integrated loudness is measured over the whole programme. Short-term uses a three-second window, momentary a 400-millisecond one. LRA — loudness range — describes how much the loudness varies across the programme, and most specs treat it as descriptive rather than a pass/fail limit.

One more distinction that catches voice people out: dialogue-gated measurement. Some specifications measure only the speech, ignoring music and effects. A mix that passes a full-programme measurement can fail a dialogue-gated one, and vice versa.

Broadcast and film targets

These are the quiet end of the scale, and they are the ones with formal standards behind them.

• EBU R 128 (Europe): −23 LUFS integrated, tolerance ±0.5 LU for pre-recorded programmes and ±1 LU for live. Maximum true peak −1 dBTP. Built on the ITU-R BS.1770 measurement algorithm.
• ATSC A/85 (United States): −24 LKFS, tolerance ±2 LU. The wider tolerance is the main practical difference from R 128.
• Netflix: −27 LKFS ±2 LU, measured dialogue-gated to ITU-R BS.1770-1 over the full programme. True peak limiters set at −2.3 dBTP or lower, with the goal of not exceeding −2 dBTP.

The Netflix number surprises people because it looks far quieter than the broadcast standards. It is not directly comparable — it is dialogue-gated, so it measures only the speech, while R 128 and A/85 measure everything. A Netflix mix and an R 128 mix can have similar perceived dialogue levels despite the four-LU gap on paper.

Streaming and podcast targets

Here the numbers are louder, and the word target means something different. These platforms normalise on playback, so the figure is not a delivery requirement — it is the level your file will be adjusted to whether you like it or not.

Spotify publishes its numbers directly, and they are worth reading carefully:

• Playback is adjusted to −14 dB LUFS, per the ITU 1770 standard
• Master to −14 LUFS integrated and keep true peak below −1 dBTP
• If your master is louder than −14 LUFS, keep true peak below −2 dBTP — louder material is more prone to distortion during lossy encoding
• Premium listeners can choose Loud (−11 LUFS), Normal (−14) or Quiet (−19)

Spotify also applies positive gain to quiet masters, but only within the available headroom — a track at −20 LUFS with a true peak of −5 dBFS gets lifted to −16 LUFS, not −14. Leaving headroom is not free; it costs you loudness you cannot get back.

The other platforms are widely reported to cluster as follows, though they publish less formally than Spotify: YouTube around −14 LUFS, Apple Music around −16, Apple Podcasts around −16. Treat these as working figures rather than specifications, and verify against the platform's own documentation before a critical delivery.

Why your loud master gets turned down anyway

Normalisation removed most of the incentive that drove the loudness war. If a platform is going to pull every track to the same perceived level, mastering ten decibels hotter than your competitors buys you nothing except the dynamic damage you did to get there.

For voice specifically, over-compression costs more than it does for music. Speech intelligibility depends on transients — the consonants that separate one word from another. Squash those and the recording gets loud and simultaneously harder to follow, which is the opposite of what a narration is for.

What to deliver when nobody specifies

Clients frequently do not state a target. Some defaults that will not embarrass you:

• Standalone voice stems for a mixer: leave them alone. Do not normalise, do not limit. Deliver at a sensible recording level with peaks well clear of clipping, and let the mix engineer place them. This is the most common mistake in remote voice delivery.
• Finished podcast or web audio: −16 LUFS integrated, true peak −1 dBTP. Sits comfortably across the podcast platforms.
• Finished audio for European broadcast: −23 LUFS, true peak −1 dBTP.
• E-learning and internal video: −16 to −20 LUFS is typical. Ask, because these are played back in wildly varying environments.

And always ask whether the client wants the loudness measured across the whole file or dialogue-gated. It is a ten-second question that prevents a re-delivery.

Recording level is not loudness

These two get conflated constantly, and the confusion produces bad recordings.

Recording level is where you set the gain going in. Loudness is a property of the finished file. You cannot fix a badly gained recording by normalising it afterwards, because normalisation raises the noise floor along with the voice.

Track peaks somewhere between −12 and −6 dBFS while recording. That leaves room for the sudden loud line nobody warned you about, and it keeps the signal comfortably above the noise floor of any competent preamp. Chasing 0 dBFS on the way in is a habit left over from 16-bit tape-era thinking; at 24-bit you have far more headroom than you need, and nothing to gain by using it.

The loudness target then gets hit at the end of the chain, after editing and any processing, using a meter rather than your ears. Ears are excellent at judging tone and terrible at judging absolute level, especially after two hours in front of the same monitors.

One consequence worth spelling out: if a client sends a spec asking for both a peak level and a LUFS target, and those two numbers conflict for your material, ask which one governs. Usually it is the loudness figure, with the peak stated as a ceiling rather than a goal. But confirm it, because occasionally a broadcaster means the opposite.

Measuring it

Any meter that implements ITU-R BS.1770 will give you the same integrated reading — that is the point of having a standard. Most DAWs now include one; the free ones include Youlean Loudness Meter and the loudnorm filter in FFmpeg, which is useful for batch checking a folder of files before you send them.

Measure the integrated value over the whole delivered file, not a section. Momentary and short-term readings are for monitoring while you work, not for reporting compliance.

If you are commissioning rather than recording, the simplest protection is to state the target in the brief. Professional voice talent will hit it without complaint — the problem is almost never capability, it is that nobody said. When you brief a project, put the loudness target and the true peak ceiling in writing alongside the script, and you will not have this conversation twice.