How does one buzz become every vowel?

In three steps, from the bottom up.

1. Breath

   Your lungs push air up the windpipe at a little more pressure than the room.

2. Buzz

   At the top of the windpipe, the vocal folds close and are blown open again and again, letting the air out in puffs: about 110 a second for an average man’s speaking voice, about 200 for a woman’s. That stream of puffs is the buzz.

3. Tube

   The buzz passes up through the throat and mouth, and out through the lips (and the nose, for sounds like “m” and “n”). That tube shapes it into a vowel.

Phoneticians call this the source–filter model: the folds are the source, the tube above them the filter. Gunnar Fant set it out in 1960, and it still underlies how speech is analysed and synthesized. The two parts are almost independent, which is why you can sing any vowel on any note.

Because it is a train of sharp puffs, not a smooth wave. A sound that repeats 110 times a second, but is not a pure sine wave, is made of a whole row of pure pitches at 110, 220, 330, 440 Hz and up, the harmonics, each a little weaker than the last. For a voice they fall by roughly 12 dB an octave. You hear the lowest as the note; the rest make it sound like a voice instead of a whistle.

Sing higher and the harmonics spread further apart; sing lower and they crowd together. But they come out of the vocal folds the same whatever vowel you are about to say. The vowel is added on the way out.

A tube rings at certain pitches, its resonances, like a bottle you blow across. Harmonics near a resonance come out loud and the rest are damped, so the tube draws a shape over the row of harmonics. The peaks of that shape are the formants, numbered from the lowest: F1, F2, F3.

Moving your tongue, jaw and lips reshapes the tube and moves the formants. Roughly, F1 rises as the jaw opens and the tongue drops, and F2 rises as the tongue moves forward. With the tongue high and forward, an average man’s tube rings at about 342 and 2,322 Hz: “ee” as in “heed”. Drop the jaw and pull the tongue back and the peaks slide to about 768 and 1,333 Hz: “ah” as in “hod”. Those are the averages Hillenbrand and colleagues measured in 1995 from 139 speakers. Plot F1 against F2 and the vowels spread into a map, which is why the first two formants are usually enough to tell them apart; listeners also use F3, how long the vowel lasts and how it glides.

Mostly the source. Singing holds the buzz at a steady pitch and moves it from note to note, often with vibrato, a regular wobble of about five or six times a second. The formants still set the vowel, which is why a choir can sing “ah” on every note of a chord.

There are limits. When a soprano sings a high note, her buzz’s harmonics are so far apart that some formants fall between them, so vowels get harder to tell apart up there, and singers adjust their mouths to keep a harmonic near the first formant. Trained singers also learn to cluster the third to fifth formants together, the “singer’s formant” near 3 kHz, which helps a voice carry over an orchestra.

That is what CHIMERA tries to do. Jameson Hodge’s free, open-source synthesizer runs in the browser and makes every sound from a simulated airway: air moving from lungs through vocal folds into a branched network of tubes for the throat, mouth and nose. It speaks typed text, sings from a piano roll and beatboxes, and in early November 2026 people made it sing barbershop and Queen. Its vowel shapes were tuned to Hillenbrand’s averages, and by its own calibration file their formants land within about 2% of their targets.

Hodge contrasts it with Vocaloid, which, he writes, “stitches together and pitch-shifts phoneme recordings.” It does not sound human yet, and he says he is “still working on improving naturalness by tuning it against recordings of real singers.” Its documents put numbers on the gap: in the project’s own test, the Whisper speech recognizer gets about 37% of its words wrong, against 2.7% for 23 real speakers. Its vocal folds follow a set pattern rather than vibrating on their own, and its “s” sounds are too weak. Held vowels are the easy part; consonants, timing and the folds are hard.

1. 1 See

   Open Vowel Lab. It starts on ee · heed: F1 342 Hz, F2 2,322 Hz. Press Hear this vowel.

2. 2 Change

   Tap ah · hod, or drag the gold dot down and to the right. F1 rises to 768 Hz and F2 falls to 1,333 Hz; the harmonics stay where they were, only their loudness changes.

3. 3 Understand

   Turn Pitch of the buzz up and down: the vowel stays “ah”. Then set Voice to Child: every formant moves up, because a shorter tube rings higher.

Vowel Lab is a formant model of an average voice: a buzz with harmonics falling 12 dB an octave, shaped by four resonances placed at measured averages. Real vowels glide, change with the sounds around them, and vary widely between speakers and accents; the averages hide that spread. The model has no nose, no breath noise and no consonants, and its sound is a classic formant synthesizer’s, robotic on purpose.

The side view of the airway is simplified and not to scale. It follows the rule of thumb that F1 tracks the jaw and F2 the tongue’s front-to-back position, which is a textbook simplification, not a measurement. CHIMERA’s figures are from its own repository: its calibration compares the formants of its static vowel shapes, not of its rendered audio, with their targets.