Versely
    Guides

    Loudness Targets by Destination, Not One Number

    Use published integrated, true-peak, and LRA figures per destination, and keep two masters when YouTube, Netflix, BBC, and podcasts disagree.

    Versely Team••9 min read

    Teams that normalise every piece to "YouTube −14 LUFS" then fail Netflix, BBC, and podcast QC are not mixing badly. They are mixing to one number as if it were a law. It is not. EBU R 128 is a published programme-loudness target of −23.0 LUFS. Netflix measures something else, with a different gate, at a different number. Apple Podcasts publishes a third. YouTube does not publish an integrated target at all. One bounce cannot satisfy those four on purpose. The job is to pick the destination, mix to its published figures, and cut a second master when the destinations disagree by more than a trim.

    This is not about why a clip sounds quiet next to other videos in a feed. That is a normalisation-direction problem, and it has its own article. This is the table you mix from, and the rule for when one file is a compromise you will regret.

    The numbers, by destination, with the document attached

    Every row below is a published spec or, where the platform has not published one, labelled as a working figure. Do not flatten them into "loudness is −14."

    Destination Integrated True peak Loudness range (LRA) What is actually being measured
    BBC / EBU broadcast (R 128) −23.0 LUFS. ±1.0 LU when the target is not practically achievable (R 128's own example is live). ±0.2 LU in QC to cover meter error. −1 dBTP in production, ±0.3 dB meter tolerance. Lower ceilings are allowed for some distribution chains (see EBU Tech 3344). Descriptor, not a hard cap. Defined in EBU Tech 3342. R 128 does not recommend LRA at all for programmes under 1 minute. Whole programme, no extra emphasis on speech, music, or effects. ITU-R BS.1770 plus EBU Tech 3341.
    EBU short-form ads and promos (R 128 s1) −23.0 LUFS, ±0.2 LU in QC. −1 dBTP, ±0.3 dB. Not applicable. Too few 3-second windows. Whole programme. Extra cap: maximum short-term loudness −18.0 LUFS (±0.2 LU in QC).
    Netflix branded near-field (Sound Mix Specifications v1.6) −27 LKFS ±2 LU, dialog-gated, ITU-R BS.1770-1, full programme. If the mix is under 15% dialogue, use programme measurement instead: −24 LKFS ±2 LU, BS.1770-3 or -4. Mix spec: peaks must not exceed −2 dB true peak. Best practice: true-peak limiter at −2.3 dBFS on deliverables. Best practice, not a fail: 5.1 and 2.0 programme LRA 4–18 LU; dialog LRA 10 LU or less. Dialog-gated 1770-1 is not the same measurement as R 128's programme loudness. LKFS and LUFS are equivalent units; the gate is not.
    Apple Podcasts (creator audio requirements) Around −16 dB LKFS, ±1 dB, ITU-R BS.1770-5. True-peak not above −1 dB FS. Precondition before encode. Not specified. Whole file. Apple notes that codecs will not fix loudness and may clip if the true-peak cap is ignored.
    YouTube Not published as an integrated LUFS target in YouTube Help. Community measurements cluster around −14 LUFS as a working figure. Not published as a dBTP cap in that Help article. The upload encoding page specifies AAC-LC / Opus / Eclipsa, 48 kHz, not a loudness number. Not published. Treat −14 as a field estimate, not a spec you can cite the way you cite R 128.

    Two arithmetic facts from that table:

    • Netflix −27 dialog-gated and R 128 −23 programme-gated are not four LU apart. They are different measurements. You cannot convert one into the other by adding a constant.
    • Apple's −16 ±1 is −17 to −15. R 128's −23 is 7 LU quieter than Apple's −16 target, and six to eight LU quieter than that whole band. A single bounce cannot sit in both windows.

    EBU R 128 s2 exists because streaming is not broadcast. It still says produce to R 128 and, unless metadata is handling device gain, stream unchanged at −23.0 LUFS. The only higher figure it names is an interim Distribution Loudness Level of −20.0 to −16.0 LUFS, for the case where the broadcaster wants to control the dynamics treatment themselves, and it tells those broadcasters to switch back to −23.0 once mobile devices have the headroom. If a client says "make it R 128 for YouTube," ask which document they mean, then mix to the one they can fail you on.

    When you keep two masters instead of one compromise

    A compromise master is a file aimed at the midpoint of two specs. −18 LUFS feels like diplomacy. It is out of spec for R 128 QC (±0.2 around −23), out of spec for Apple Podcasts (±1 around −16), and meaningless to Netflix's dialog-gated meter. You have produced a file nobody asked for.

    Keep two masters when any of these is true:

    1. The destinations use different gates. Netflix dialog-gated 1770-1 versus R 128 programme-gated is the split that bites. Mix once, measure twice, trim the programme master from a copy. Do not aim between.
    2. The integrated windows do not overlap. Apple −16 ±1 and R 128 −23 do not overlap. YouTube's working −14 and R 128 −23 do not overlap. Two bounces. Same mix, different output gain and limiter ceiling.
    3. True-peak ceilings disagree. R 128 production is −1 dBTP. Netflix mix spec is −2 dBTP, with a recommended limiter at −2.3. Limit the Netflix bounce to their ceiling; do not loosen the BBC bounce to match.
    4. LRA is named on one sheet and forbidden on the other. Netflix's best-practice band is 4–18 LU programme, dialog LRA 10 LU or less. R 128 s1 says LRA is not applicable on short-form. Do not squash a 20-second ad to hit 4–18.

    What two masters looks like in the session:

    • Master A, broadcast / BBC / EBU: sequence at 48 kHz / 24-bit, integrated −23.0 LUFS programme, true peak ≤ −1 dBTP, short-term ≤ −18.0 if the piece is an ad or promo under R 128 s1. WAV (or broadcast MXF, if that is the ticket).
    • Master B, podcast: same mix, output gain up so integrated sits at −16 LKFS ±1, true peak ≤ −1 dB FS, then encode. Apple wants the loudness and true-peak pass before AAC or MP3.
    • Master C, Netflix near-field: dialog-gated 1770-1 at −27 LKFS ±2, true-peak limiter at −2.3. Under 15% dialogue, switch to −24 LKFS ±2 programme (1770-3/4). Different meter mode, not a trim of Master A.
    • Social / YouTube working copy: trim a copy toward the community −14 figure, true peak under −1 dBTP. Label it a working copy. Do not file it as "the R 128 master."

    The mix (voice versus bed, EQ, fold-down) is shared. The output gain, the limiter ceiling, and the meter mode are not. Ducking the bed under a voiceover still happens once, on the mix. Loudness is the last pass, per destination.

    Measure the right gate, then sign off per destination

    A loudness meter with the wrong gate will tell you you passed while QC fails you. Set the mode to the document.

    • R 128 / BBC: programme loudness, relative gate, EBU Tech 3341 "EBU Mode." Whole file, no dialogue emphasis. That is R 128 item (l).
    • Netflix branded: dialog-gated, ITU-R BS.1770-1, full programme. Switch to programme-gated 1770-3/4 only when dialogue is under 15%. Use a meter the spec has tested, in that mode.
    • Apple Podcasts: ITU-R BS.1770-5, whole file, −16 ±1, true peak ≤ −1 dB FS, before encode.

    Versely's editor does not replace that meter. The audio group gives you music_volume and original_volume on a static mix, which is relative level, not a BS.1770 integrator. Get the balance in the editor (tool by tool), bounce a 48 kHz / 24-bit WAV, then measure. Check the video before you publish is the picture-and-captions pass; loudness still needs a meter that names BS.1770.

    A podcast episode is its own destination even when the same footage will become clips. Long-form podcast production should be mastered to Apple's −16 ±1 and the clips cut from a copy trimmed for the feed they will live on.

    The 480p preview pass is free, with a short per-user cooldown, and it will tell you whether the balance survived. It will not tell you whether integrated loudness is −23.0 or −16. That number comes off a BS.1770 meter on the WAV, per destination, before anyone presses encode.

    FAQ

    Can I just mix to −14 and let Netflix or the BBC normalise me?

    Netflix says it normalises deliveries so titles play at a consistent level, and that it does not compress or limit to "fix" a mix. A file that misses −27 LKFS ±2 dialog-gated is still out of spec. BBC / R 128 QC is a measured −23.0, not a hope that the chain will rescue you. Mix to the sheet you will be failed against.

    Why doesn't a −23 mix just get turned up on YouTube?

    Platform normalisers are not required to boost quiet files as aggressively as they turn down hot ones, and YouTube has not published the integrator you would be aiming at. A −23 file in a feed of −14-ish working copies will often just sit quieter. That is the quiet-video problem, not a reason to throw away an R 128 master you need for broadcast. Keep both.

    Do I need an LRA number on a 20-second ad?

    Not under R 128 s1. Loudness Range is not applicable there; the short-term cap (−18.0 LUFS) is the dynamics control. Netflix's 4–18 LU band is best practice for programme-length near-field mixes, not a 20-second promo. Put LRA on the spec sheet only when the destination named it.

    Is LKFS different from LUFS?

    R 128's footnote: LUFS is equivalent to LKFS as used in ITU-R BS.1770. The EBU prefers LUFS. A Netflix LKFS figure and an EBU LUFS figure are the same unit. They are not the same measurement if one is dialog-gated 1770-1 and the other is programme-gated R 128. Quote the document, not just the letters.