The short answer
Your voice is a buzz from your vocal folds played through the tube made by your throat and mouth. That tube lets some frequencies out louder than others, and those loud ones are your formants. Change the shape of the tube and you change where they sit.
A smile is a change to that shape. Pulling your lips back shortens the tube, and a shorter tube resonates higher, the same way a shorter pipe plays a higher note. Nothing about your pitch has to move for a listener to notice.
What actually moves, and what does not
That measurement is from a study at the University of York on a corpus of spontaneous speech, which matters because most earlier work used scripted recordings where people know what they are performing. The headline there is the third formant: a difference of 53 Hz, consistent enough to be the reliable marker of a smiled voice.
The second formant going the other way is the part worth keeping, because it is the part that gets dropped. A smile is not every resonance sliding up together. It is a specific reshaping with a specific signature, which is why it sounds like a smile rather than like helium, where every resonance really does move the same way at once.
Your face answers before you decide to
A team at IRCAM and CNRS went further and worked out the exact filter. They used reverse correlation to find what you have to do to a recording to make listeners call it smiled, and the answer was a shift of the first two formants plus a boost around the third. Their algorithm runs both ways: a smile mode that moves the formants up, and an unsmile mode that moves them down.
Then they put electrodes on listeners' faces. The muscle that pulls a smile responded to the smiled recordings, and the frowning muscle did less, both with statistically significant differences. The part that should bother you in a good way: the smile muscle responded to the acoustic cue even when the listener had not noticed the recording was manipulated at all.
Why this is useful and not just strange
It is a change you can make with no gear, in the middle of a sentence, and it reaches the other end. On a call or a stream, where the other person has no face to read and a compressed signal to listen through, the shape of your mouth is one of the few expressive channels that survives the trip.
It also explains an old piece of advice that always sounded like superstition. Call centers have told staff to smile while they talk for decades, and the instruction turns out to be acoustic rather than motivational: the posture is audible whether or not the feeling behind it is real.
If you want the rest of what reaches the other end of a call, the physical and chain fixes are in sound better on video calls.
The same mechanism, on a dial
Everything above is formant movement, which is the thing a formant shifter does deliberately. The difference is scope: a smile reshapes one end of your tube and moves the resonances by tens of hertz, while a formant dial moves all of them together by a chosen amount.
It cuts both ways. A deep voice effect reads as a bigger body rather than a sad one, and why no single dial on any effects rack is labeled smile: the acoustic signature is a particular pattern across three formants rather than one slider's worth of movement. The dial moves size. Your face moves something narrower.
The takeaway
Smiling shortens your vocal tract, which raises the resonances that give your voice its character. The third formant is the reliable marker, 53 Hz higher in smiling speech measured on real conversation, while the second moves the other way. Listeners pick it out with no picture, and their own smile muscles respond to it even when they never notice the sound was changed.
Patchd is a Windows audio mixer with a full effects rack on every channel, including the formant tools this page keeps pointing at. It is in open beta now, and if you clear the terms your invitation arrives straight away.