Versely
    Guides

    Stereo Fold-Down That Keeps Dialogue

    Fold 5.1 and Atmos beds with centre-channel priority, discard the LFE, and check the result on a phone speaker so the words still survive.

    Versely Team••9 min read

    A generated music bed that sounds huge in headphones can erase the voice the moment a phone sums it to one speaker. The bed was mixed as a 5.1, an Atmos-style object dump, or a stereo file so wide and out of phase that the centre is empty. Fold-down is not "make it quieter." It is "put the words in the middle, throw away the channel a phone cannot play, and listen to the result the way the audience will."

    Most short-form video is heard on a single small speaker. Stereo is already a fold. 5.1 and Atmos are two folds. If you do not do those folds on purpose, the device does them with a default that does not know your dialogue lives in the centre.

    Where the words go when you fold

    Dialogue in a proper surround mix sits in the centre channel. That is the point of having a centre: the words stay in the middle of the picture when you sit off-axis. A Lo/Ro stereo fold that forgets the centre, or that buries it under unfiltered LFE and hot surrounds, is how a mix that passed in the room fails in a kitchen.

    Netflix's near-field 2.0 recipe, for the case where you are folding a 5.1 to Lo/Ro, is explicit:

    • Lower the centre by 3 dB and put it in both left and right.
    • Lower surrounds by at least 3 dB and fold left-surround into left, right-surround into right.
    • LFE is optional, and if you include it you drop it 8 to 12 dB, low-pass at 200 Hz or lower, and put it in both legs.
    • All of those trims happen on the 5.1 channels before the fold, not after.

    The same document's best-practice note is the one that matters for phones: ITU-R BS.775-1, 5.1 to stereo, with the LFE muted, is the international standard. Netflix allows LFE in the 2.0 as an option. For a phone speaker, mute it. A 3-inch driver cannot play an LFE channel. What it can do is turn that energy into mud under the voice.

    Dolby's default Lo/Ro coefficients match the same centre-and-surround picture:

    Lo = L + (−3 dB × C) + (−3 dB × Ls)

    Ro = R + (−3 dB × C) + (−3 dB × Rs)

    Lt/Rt (the Dolby surround matrix) is the other stereo fold. Netflix prefers Lo/Ro and requires the 2.0 to be mono-compatible. For anything that will be heard on a phone, Lo/Ro is the fold. Lt/Rt can put dialogue into a matrix that a cheap downmix to mono will notch.

    Generated output rarely arrives as a labelled 5.1 printmaster. What you actually get, in descending order of how often it wrecks a fold:

    1. A stereo master so wide it is almost decorrelated. Text-to-music tracks are mixed as records, not as beds. Hard-panned synths, chorus, and "cinematic" stereoisers put energy in L and R that cancels when you sum to mono. The centre, where the voice will sit, goes hollow.
    2. Native-audio video with source-anchored pans. A model that moves sound with the camera can put the line that matters off to one side. Headphones: dramatic. Phone speaker: the line is gone.
    3. A 5.1-tagged or Atmos-wrapped file from a music or trailer pipeline that assumed a living room. Default device fold, no centre priority, LFE included.

    Treat all three as the same job: produce a Lo/Ro (or a centred stereo) that still has the words when you hit Mono.

    Centre-channel priority, and why you discard the LFE

    Centre-channel priority means the dialogue path is the one you protect, and everything else is trimmed relative to it. In a discrete 5.1 that is literal: C is the dialogue stem, you put it in both legs at −3 dB, and you do not let Ls/Rs or LFE out-loud it. In a generated stereo file there is no C bus. You make one:

    1. Put the voice on its own track, centred, dry. If the voice is baked into a native-audio clip, isolate it or replace it with a centred voiceover. A voice that only exists as a hard pan will not survive a phone.
    2. Fold the bed, not the voice. If the bed is 5.1, use the Netflix Lo/Ro trims with LFE muted. If the bed is stereo, sum a copy to mono and listen. If the mono copy loses the hook, the bed is too wide; collapse it (mid/side, pull the side down 3–6 dB) before you mix it under the voice.
    3. Mute LFE for anything that will be heard on a phone, laptop, or TV speaker. LFE is a band-limited effects channel, not "the bass." Folding it in at full level is how a trailer hit becomes a fart under a product line. Netflix's own default international fold mutes it. Follow that unless a 5.1 living-room deliverable is the actual destination, and even then low-pass at 200 Hz and drop it 8–12 dB.

    Do not "make up" a centre by boosting 2–4 kHz on the whole stereo mix. That hypes hats and synths as much as consonants. Priority means the channel the words live on, not a presence EQ on the sum.

    A fold-down you can check, then the phone test

    Do the fold as a bounce, not as a monitoring button you forget to export.

    If you have discrete 5.1 (L, R, C, LFE, Ls, Rs):

    1. Mute LFE.
    2. C into L and R at −3 dB.
    3. Ls into L at −3 dB (or lower if the surrounds are music).
    4. Rs into R at −3 dB.
    5. L and R at unity.
    6. Bounce Lo/Ro, 48 kHz / 24-bit.
    7. Sum that bounce to mono and listen. If the words dip, the surrounds or a wide L/R bed are fighting C. Lower surrounds another 3 dB and bounce again.

    If you have stereo only (the common generate):

    1. Centre the voice. If it is a native-audio take with the line off-axis, re-centre it or re-read it.
    2. Mid/side the bed. Pull the side down until a mono sum still has the musical idea, then stop. You are not trying to make the bed mono. You are trying to make it survive mono.
    3. Mix the bed under the voice as a static relationship. Ducking the music is the Versely-side volume step (attach_audio_to_video in mix mode). Fold first, duck second. Ducking a bed that cancels in mono just gives you a quieter hole.

    Then the check that actually decides whether this ships. It is the same phone-speaker check the spatial-audio piece uses, aimed at words rather than pans:

    1. Play the Lo/Ro (or the collapsed stereo) on a phone speaker at arm's length, in a room with ordinary background noise. Not on the sofa with headphones. Not in the NLE.
    2. Confirm you can follow the line without filling in words you already know. If you wrote the script, you are a bad listener for this; play it to someone who did not.
    3. If the line is gone, the fix is not more width and not more LFE. It is more centre: voice up and centred, bed narrower, LFE still out.

    Do this on a preview, not on a paid master. The editor's 480p preview pass is free, with a short per-user cooldown, and audio comes through the downscale; the final export is charged once. A preview answers "do the words survive a speaker." It does not answer "did I mute the LFE in a 5.1 printmaster." That one you check on the bounce meters.

    Adding music to a video after a fold is the right order: fold (or collapse) the bed, then attach it under the voice. Attaching a 5.1 or a hyper-wide stereo and hoping the platform will fold kindly is how a model with audio loses the line on a commute.

    FAQ

    Should I ever include the LFE in a stereo fold?

    For a living-room 2.0 that Netflix might stream, their spec allows it at −8 to −12 dB, low-passed at 200 Hz or below, in both legs, and the resulting 2.0 still has to be clean. For phones, laptops, and anything you will check on a speaker in your hand: no. Mute it. The international default fold (BS.775-1 with LFE muted) is the one that matches that device.

    My generate is stereo. Do I still need a "fold-down"?

    Yes, because the phone will sum it. A fold-down of stereo is a mono-compatibility check plus a width trim. If the mono sum keeps the words and the idea of the bed, you are done. If it does not, the stereo image is the problem, even though no 5.1 was involved.

    Can I leave the surround mix alone and let YouTube or the TV fold it?

    You can, and you will not like the default. YouTube will accept stereo or stereo+5.1. It will not protect your centre channel. A fold you do not listen to is a fold you do not control. Bounce Lo/Ro, listen on a phone, then upload the 2.0 you actually signed off.

    Does collapsing the bed make it sound small in headphones?

    A little, on purpose. A bed that is huge in headphones and empty in the centre is a bed that fails the majority playback path. Pull the side down until the phone test passes, then stop. You can keep a wider 5.1 or Atmos as a second master for the destinations that will play it. Do not make the phone suffer so the headphones feel expensive.