A Pitch Sound Disappears in Mono. Fix the Mix or Change the Effect?

There's always one sound in a pitch mix doing the actual selling. A fizz, a single struck note, a breath before the line. On headphones it's the best ten seconds on the board. On the client's smart speaker it isn't there at all, and nothing else is wrong — which is the part that gets under your skin. The spot plays. The voice is fine. The hit is fine. Just not that.

You have one question to answer before any repair makes sense: is the sound quiet, is it routed somewhere it doesn't reach, or is it cancelling when the channels combine? From the outside all three look identical — one element missing under one playback condition — and they have almost nothing in common as fixes. A fader cures one of them and does precisely nothing for another.

After that the fork is real. Keep the effect and rebuild the mix around it, or change the effect for one that survives the fold. There's usually a third path between them, and it's often the right one: let the effect do the spectacle and let a different layer carry the information. Those are separate jobs. Nothing says one sound has to do both.

Some of what follows is arithmetic, which is exact and checkable on paper. Some of it is a constructed example, because I can't hear your session and I'm not going to pretend otherwise. I'll keep the two clearly apart.

Reproduce the disappearance before you explain it

Start with a comparison, not a fix. Three listens of the same passage, matched as closely as you can manage:

• the full stereo mix on the route it was mixed on;
• each channel on its own;
• a true sum of both channels.

Change nothing else. Don't rebalance, don't re-process, don't "just try" a fix in the middle of the comparison. You're establishing a fact about one element, and every other change is noise in that evidence.

Then get clear on what "mono" you're actually making, because the word covers several different operations. A DAW mono button that sums L+R is a sum. A mono-maker plug-in that nulls the side channel gives you (L+R)/2 — identical content sitting 6 dB lower, which is fine for asking is the element present and useless for asking is the level right. A mono button that people assume is a sum but which really solos one channel tests nothing about summation at all. A single physical speaker receives a sum in the air. A stereo Bluetooth pair receives no combination whatsoever and will never reproduce this problem. And a half-seated plug kills one channel outright, along with everything on it — that's a routing failure, and it looks nothing like cancellation.

Match loudness before you judge any of it. In a DAW monitor path, folding to mono commonly makes the program louder, because correlated content now feeds both speakers with the same signal, while whatever was anti-phase gets nothing from that at all. The fold can sound bigger, punchier, more solid — and the fizz is still gone. Louder is not fixed, and a fold isn't a level-matched comparison until you make it one.

Decide early which routes you're actually testing for. For a pitch, that's usually the meeting-room laptop speakers, a phone held up in a kitchen, a smart speaker, and whatever the exec is wearing on the train. That's the bench. You can't test every playback path and you shouldn't promise the mix is equivalent on all of them. You can give a specific account of what happens on the routes that matter.

Quiet information, missing information, cancelled information

Take a single mono source and place it four ways across two channels. Call the source s.

Those are linear amplitudes. Doubling amplitude is +6 dB, so a single copy at s is −6 dB against the in-phase duplicate's 2s. The half-inverted copy sums to 0.5s, a quarter of that duplicate, which is −12 dB against it. That's the whole scale of the problem in one row of a table.

Two things the table doesn't say. First, it describes the mono bus, not the mix. Real sounds have spectra, so real problems are usually partial and frequency-dependent rather than the clean cancellation in the last row. Second, it describes one element in isolation, and what matters is always that element's fold level relative to everything else that folds with it.

Still, two useful categories fall out of it.

Partial. L = s, R = −0.5s sums to 0.5s. The element is present in the fold, just 12 dB below where a correlated copy would have put it. That's quiet information, and level might genuinely be the right repair — but only now, knowing it's partial, and knowing that raising it raises the comb peaks along with it.

Exact. L = s, R = −s sums to nothing. No fader position recovers it. Multiply zero by any gain you like. Full cancellation is the one case where a mix-level fix isn't available at all and you have to change something structural: the routing, the effect, or the source itself.

The reason this is hard to hear is that a resolved, well-balanced stereo mix is not a preview of its fold. It's a different mix. The stereo version tells you the idea works. It tells you nothing about whether the idea survives summation.

Follow the path before you change the character

Work in order: source, track, bus, output. The usual suspects in a pitch session are duplicated tracks (someone made a copy to widen something and never deleted the original), sends feeding a reverb or parallel compressor you'd forgotten about, any widening or stereoizing stage anywhere in the chain, and polarity buttons left on from an earlier fix that made something else better.

Adobe's documentation for Audition's multitrack routing describes track signals being routed through buses and outputs, and lists polarity inversion as an available operation. I haven't rechecked that documentation for this piece. It's useful in a narrow way — it confirms the controls are ordinary and available. It tells you nothing about your mix, and an available operation is not a repair.

While you're in there, keep two words apart, because confusing them costs time.

Polarity is a flip. Multiply the whole signal by −1. Identical at 20 Hz and at 20 kHz.

Phase is an offset in time, and its effect is frequency-dependent. Delay one channel by 9 ms and the two channels disagree by a different angle at every frequency.

That distinction is practical. A comb built from a delay cannot be fixed "by flipping the polarity" in any general way; a flip just relocates which frequencies notch. If someone tells you the problem is phase and the answer is invert it, one of you is using the wrong word.

A diagnostic that earns its keep: solo the element, sum its two channels, and listen. Then invert polarity on one channel of that element only and listen to the sum again.

• If it gets louder, the element has anti-phase content, partial or full. You've found the mechanism.
• If it gets quieter, it was correlated and mono-safe, and the problem is elsewhere — level, routing, or your test setup.

Meters earn their keep too, up to a point. A correlation meter pinned near −1, or a vectorscope showing a thin horizontal stroke instead of a rounded cloud, is real evidence that a layer is anti-phase. It is not evidence about whether the pitch works without that layer. That judgment has no readout.

Three repairs, and what each one costs

Now the constructed part. There is no Coldacre. It's a sparkling water I made up, and I haven't heard a second of this spot.

The pitch is thirty seconds and sound-led. Room tone of a lake at dawn. A hand lifting a bottle out of the water. One bright cap crack. The pour, and then the fizz. The fizz is the hero — it's what carries cold arriving. The voice reads the line over it.

The fizz was built from a mono recording of carbonation, duplicated, the copies hard-panned left and right, with polarity inverted on the right copy. In headphones it's remarkable: the fizz stops having a location and seems to happen around the listener. In the fold, L + R = 0. The cap crack is centred and survives. The voice is centred and survives. The fizz is gone. The failure is brutally specific, which is the good news — you know exactly what you're repairing.

Two traps first, because both look like fixes.

Trap one: add a centred copy of the same fizz. Write the anti-phase layer as a·s and the centred copy as b·s.

L = (b + a) · s
R = (b − a) · s
L + R = 2b · s

The fold hears only the centred copy, which is what you wanted. Now set b equal to a, and L = 2s while R = 0. You haven't anchored a wide sound. You've panned a mono sound hard into one channel. The mid and the side have to be different signals, or the mid simply pans the source. This is the single most useful thing to know about anti-phase widening, and it's why the obvious repair fails.

Trap two: dial the inversion down until the fold is acceptable. The correlation of a source split as (b+a) and (b−a) doesn't slide gently from −1 toward +1. It flips sign. There's no comfortable midpoint where the sound is both wide and centred; you cross a discontinuity. If you're hearing a smooth transition, you're probably listening to level, not correlation.

Repair A — change the technique, keep the width

Replace the inverted duplicate with something whose fold retains energy.

An inter-channel time offset is the cheapest swap: left leads by 9 ms, right lags by 9 ms, 18 ms of total difference. In the fold the two copies sum into a comb with notches every 55.6 Hz. On broadband fizz that reads as a mild hollowing, not a disappearance — the information survives. On a pitched element the same comb is exposed, and a bell or a voice through it sounds obviously filtered. Fine for noise, risky for pitch.

Genuinely decorrelated stereo is usually better. Two different fizz recordings, or one recording processed differently on each side. Uncorrelated material sums to roughly +3 dB instead of +6 and doesn't cancel at all. You lose the no-centre sensation and keep a wide, present fizz with no comb.

Cost: the specific inside the headphones quality is exactly what leaves. What comes back is a fizz that has a place. If the place is the problem, this is the wrong repair.

Repair B — fix the mix, carry the information on another layer

Keep a reduced anti-phase layer for the far edges. Add a separate, mono-safe fizz layer in the middle carrying the same idea — different recording, different processing, doesn't matter, as long as it isn't a copy of the same source. Because it's a different signal, adding it doesn't pan anything. The fold has the mid layer. The stereo version still has some of the spectacle.

Cost: two layers to blend and keep from reading as two sounds. The folded version is the mid layer alone, which isn't the fizz the designer fell for. And you now have to decide how much of the original effect is still enough to justify the extra machinery.

Repair C — change the source, centre it

Drop the widening. The fizz becomes a mono element and gets its size from the arrangement instead: a quieter bed immediately before it so it arrives into contrast, a mono-safe reverb return, a low layer doing the room, an automation move across the pour.

Cost: everywhere at once is gone. The fizz has a location again.

Benefit: for this element, the fold and the stereo version carry the same information. Nothing to reconcile. It's the cheapest and fastest option, and for a lot of pitches it's the right one.

Which repair is correct depends entirely on what the fizz was for. If the idea was "it's all around you," the width is the idea and Repair A or B is the honest fix. If the idea was "cold arrives," the width was decoration, and Repair C is cheaper and probably better. Don't assume the widest stereo version deserves rescuing at any cost. The stereo mix isn't the master and the fold isn't the compromise. Both are deliverables, and both have to carry the idea.

Check the whole moment, not the component

Listen to the repaired passage in context, on the intended route, and in the controlled fold. Then listen to the seconds on either side, because that's where repairs do their damage.

Three specific risks. A fizz raised to survive the fold sits somewhere around 4–10 kHz, which is exactly where consonants and sibilance live — it is entirely possible to fix the fold and lose the line. If the cap crack used to land into a wall of fizz and now lands into a smaller one, you've fixed mono and weakened the arrival; that isn't automatically a bad trade, but it is a trade, and somebody should make it deliberately. And check for new routing damage: a fresh bus, a duplicated track now feeding the master twice, a fix that existed on your monitor path and didn't make it into the bounce.

Keep the untouched version. You'll want it for a stereo-only deliverable, and you'll want it the next time somebody says the repair sounds different.

Then write the compromise down in a sentence another person can read. Not "now mono-compatible" — you can't promise that, and it isn't true of every playback route. Something closer to: in a summed fold the fizz sits under the voice by a few dB and reads as a layer rather than a moment; in stereo it's wide with a soft centre; we traded the no-centre sensation for the fold.

That sentence is the end of the job. Not a mix that fails nowhere, which doesn't exist, but a mix whose failure you measured, chose, and can explain to the person who has to approve it — a repaired route or a changed effect, with the width you gave up said out loud. The fizz has to be there on the smart speaker. Which version of the fizz arrives is a decision, and it's better made on purpose than discovered in the meeting.

Frequently asked questions

What is the first question to answer when a pitch sound disappears in mono?

Is the sound quiet, routed somewhere it does not reach, or cancelling when the channels combine? From the outside all three look identical—one element missing under one playback condition—but they have different fixes. A fader cures one and does nothing for another.

What does a mono sum show for centred, hard-left, half-inverted, and fully inverted placements?

A centred duplicate sums to 2s, the reference. Hard left sums to s, which is -6 dB. A half-inverted copy sums to 0.5s, a quarter of the duplicate, or -12 dB. A fully inverted copy sums to 0 and is gone. These are linear amplitudes; doubling amplitude is +6 dB.

Why can adding a centred copy of the same anti-phase fizz fail?

If the centred copy and the anti-phase layer are the same signal, setting the centred copy equal to the anti-phase layer makes L = 2s and R = 0. You have not anchored a wide sound; you have panned a mono sound hard into one channel. The mid and side have to be different signals.

What are the three repair paths and their main costs?

Repair A changes technique while keeping width, such as an inter-channel time offset or genuinely decorrelated stereo; it loses the inside-the-headphones quality and can comb pitched elements. Repair B keeps a reduced anti-phase layer and adds a separate mono-safe middle layer; it costs two layers to blend and the folded version is the middle layer alone. Repair C centres the source and builds size in the arrangement; everywhere-at-once is gone.

What should be checked after a repair?

Check the whole moment in context, on the intended route, and in a controlled fold, then listen to the seconds on either side. Watch for a raised fizz in the 4–10 kHz range masking consonants or sibilance, a cap crack that now lands into a smaller fizz, and new routing damage such as a fresh bus, a duplicated track feeding the master twice, or a fix that existed only on the monitor path. Keep the untouched version and write the compromise in a sentence another person can read.