DeliveryQC Beta
Netflix delivery reference
How to pass Netflix audio QC
Failing a QC check costs you time. You need to revisit your mix, find the problem, and submit it again. It's a hassle. There are quite a few things that can trip up a delivery, but loudness and true-peak levels need not be one of them.
Netflix doesn't publish one set of audio specs. What you have to hit depends on which deliverable you're making. And the loudness figure — whichever spec you land on — isn't the one a "standard" loudness meter reports.
Let's step through it...
Note: Netflix revises its specifications periodically, and your delivery paperwork takes precedence.
First: which mix are you delivering?
Netflix's audio requirements split by deliverable. Before anything else, work out which of these you're targeting.
Near Field mix
Your final mix, for home viewing rather than a cinema. Delivered as Atmos, 5.1 Surround or 2.0 Stereo.
Theatrical mix
Mixed for cinema playback. The near-field numbers don't apply — work from the theatrical spec.
M&E
Your mix with the dialog removed, for dubbing. No dialog means nothing for a dialog gate to measure.
From here on, this page is about the loudness and true-peak requirements for Near Field mixes.
Then: which Near Field mix?
There are three variants, and they don't carry identical requirements. The loudness target is the same for all three; where they differ is in what gets measured, and in how hard the true-peak limit is.
| Asset | Loudness | True peak | Channels |
|---|---|---|---|
| Near Field 5.1 Surround Mix | -27 LKFS ±2, dialog-gated | ≤ -2 dBTP — hard | 5.1 |
| Near Field 2.0 Stereo Mix | -27 LKFS ±2, dialog-gated | ≤ -2 dBTP — hard | 2.0 |
| Near Field Atmos Mix | -27 LKFS ±2, on the 5.1 re-render | -2 dBTP — advisory | ADM BWAV |
All three are delivered at 48 kHz / 24-bit.
Two things are worth pausing on before you go further... First, the loudness measurements are dialog-gated, which is different from what a standard loudness meter reports. More on that below. Second, dBTP is true peak, which reconstructs what the waveform does between samples; it reads higher than a sample peak meter, often by a dB or more.
What “dialog-gated” actually means
This is the part worth understanding properly, because it explains why your meter and Netflix's can disagree by several dB on the same file.
A normal loudness meter measures everything above a level threshold: dialog, music, explosions, room tone. But when people watch a film or TV show, they typically adjust their volume based on the dialog (the "anchor" of the mix).
Dialog-gated loudness measures only the moments where speech is present. A classifier decides, moment by moment, whether there's dialog; the meter counts only those moments.
That tracks the thing your viewer actually sets their volume by: the dialog. An action sequence with no talking stops dragging your reading around.
The 15% threshold
What if your program barely has any dialog? A dialog-gated measurement only means something if there's enough speech to measure.
On a concert film, for example, the gate has little to work with — and the number describes a few scattered seconds rather than your program.
Netflix handles that with a speech-content threshold. Where dialog accounts for less than 15% of the program, loudness is evaluated as best practice against a program-loudness target of -24 LUFS ±2 instead of the dialog-gated -27.
The gap isn't arbitrary. On a mix whose program loudness is -24 LUFS, the dialog-gated loudness will typically sit around -27 LKFS.
Best practices worth following
None of these are enforced, but they're what separates a mix that passes from one that plays well.
- Loudness range roughly 4-18 LU. Loudness range (LRA) is the spread between your quietest and loudest passages. Under 4 suggests a mix squashed flat; over 18 suggests one that will be hard to follow on a phone, or in a noisy room. Neither fails — both are a prompt to listen again.
- Dialog loudness range at or under 10 LU. Similar to LRA, but for dialog. If the range is too wide, it may indicate that some dialog will be too soft to follow.
- Keep the true-peak headroom. Netflix recommends a -2.3 dBTP limiter on everything you deliver, Atmos beds and objects included.
- Check the stereo downmix. A large share of viewing is on stereo devices, so it's worth reviewing if time allows.
Worth checking before you deliver
- The gate. Is your reading dialog-gated or whole-program?
- Atmos. Measured on the 5.1 re-render, not the bed.
- Peak type. True peak, not sample peak.
- Target wiggle room. Land as close to the loudness target as you can. Then if there's a slight difference between your meter and Netflix's, you won't fail QC because you'll still be within the pass threshold.
Check a file against these specs
Know where your mix lands before you send it. DeliveryQC checks if your mix complies with Netflix's specs. If it doesn't, it tells you exactly why. Your audio never leaves your browser, and it's free!
Open DeliveryQCSources
Figures are taken from Netflix's own delivery specifications, which live on the Netflix studio partner portal and require partner access:
- Non-branded delivery specifications
- Branded audio mastering & archive specification
- Branded sound mix specification & best practices — the source for the advisory figures in “Best practices worth following”
Netflix is a trademark of Netflix, Inc. DeliveryQC is an independent tool, not affiliated with, endorsed by or certified by Netflix. The published documentation is authoritative.