How Pink Trombone Works: Vocal Tract Physics & Speech Synthesis

Explore the acoustic physics inside Pink Trombone: source-filter theory, Rosenberg glottal pulses, the 44-segment Kelly-Lochbaum waveguide, formants, and nasal coupling.

Updated 2026-11-21

Unlike modern AI voice generators or sampler choirs that stitch together recorded human audio, Pink Trombone synthesizes speech from first principles. Every vowel, hum, hiss, and squeak is computed in real time from a mathematical model of air waves bouncing inside a human throat, mouth, and nose.

Understanding how this articulatory speech synthesizer works makes it much easier to control — and reveals why the human voice behaves like a flexible brass instrument (hence Neil Thapen’s name, Pink Trombone).

The source–filter model of human speech

Acoustic phonetics divides human voice production into two coupled stages:

  1. The Source (Glottis & Turbulence): Air pushed from the lungs vibrates the vocal folds to create a harmonically rich buzz (for voiced sounds like vowels and [z]), or rushes through a narrow gap to create broadband hiss (for voiceless fricatives like [s] and [ʃ]).
  2. The Filter (Vocal & Nasal Tract): That raw excitation travels through the pharynx, oral cavity, and optional nasal branch. Because the tube reflects sound waves back and forth, certain resonant frequencies — called formants — are amplified while others cancel out.
Stage in biologyEquivalent in Pink TrombonePrimary acoustic role
Vocal folds (larynx)Rosenberg-style glottal pulse generatorSets pitch (F0), harmonic brightness (tenseness), and vibrato
Airflow turbulenceFiltered white-noise excitationProduces aspiration breathiness and fricative hiss ([s], [ʃ], [f])
Pharynx & oral cavity44-segment Kelly-Lochbaum waveguideShapes formant peaks (F1, F2, F3) via cross-sectional area
Velum & nasal cavityBranching 28-segment nasal waveguideAdds low-frequency nasal murmur and anti-formants ([m], [n], [ŋ])
Lip radiationHigh-pass radiation filter at the mouth exitModels how sound radiates from open lips into room air

Inside the glottis: Rosenberg glottal waveform

A real vocal fold does not vibrate like a pure sine wave. Instead, the folds snap shut rapidly once per pitch cycle, creating a sharp asymmetric pulse packed with upper harmonics.

Pink Trombone models this using a smooth parametric curve inspired by Rosenberg’s (1971) glottal pulse:

  • Pitch (F0): Sets how many times per second the glottal cycle repeats (typically ~85–180 Hz for low voices, ~165–300+ Hz for higher voices).
  • Tenseness: Controls the ratio of open phase to closed phase and the sharpness of vocal-fold closure. Low tenseness produces a soft, breathy whisper with strong aspiration noise; high tenseness produces a bright, brassy tone rich in high harmonics.
  • Loudness & jitter: Small organic micro-variations ("auto-wobble") mimic natural human pitch drift so sustained tones sound alive rather than robotic.

The 44-segment Kelly-Lochbaum digital waveguide

The heart of Pink Trombone is a 1D Kelly-Lochbaum acoustic tube model originally introduced in 1962 for speech synthesis. Imagine slicing the vocal tract from the larynx (segment 0) to the lips (segment 43) into 44 short cylindrical segments, each with a dynamically changing diameter:

  • At every sample tick, a forward-traveling pressure wave moves toward the lips while a backward-traveling reflection wave moves back toward the glottis.
  • Wherever two neighboring segments have different cross-sectional areas (Aᵢ and Aᵢ₊₁), a reflection coefficient kᵢ = (Aᵢ − Aᵢ₊₁) / (Aᵢ + Aᵢ₊₁) determines how much acoustic energy bounces back and how much passes through.
  • Dragging the tongue in the Pink Trombone canvas smoothly alters the diameters of segments in the throat and mouth, instantly updating all 44 reflection junctions at audio rate.

Why a 3-way junction creates nasal consonants

At roughly segment 17 (the velum / soft palate), Pink Trombone adds a three-way scattering junction connecting the pharynx, the front oral cavity, and a separate 28-segment nasal tube.

  • When the velum is closed, the nasal branch diameter drops near zero and all sound exits through the mouth (oral vowels and stops).
  • When the velum opens and you seal the lips ([m]) or alveolar ridge ([n]), the oral cavity acts as a closed side-branch resonator while sound radiates out the nostrils — producing authentic nasal formants and anti-resonances!

Formants (F1 and F2): how tube shape becomes vowels

Every vowel is recognized by the human ear primarily through its first two resonant peaks, F1 and F2:

VowelApproximate F1Approximate F2Physical cause inside the waveguide
[i] ("ee")~280 Hz (low)~2,300 Hz (high)Tight front mouth channel + wide back pharynx cavity
[ɑ] ("ah")~730 Hz (high)~1,100 Hz (low-mid)Narrow pharynx + wide open horn-shaped front mouth
[u] ("oo")~300 Hz (low)~870 Hz (very low)High-back tongue pinch + narrowed, elongated lip opening
[æ] ("a")~660 Hz (high)~1,720 Hz (mid-high)Low front tongue opens the front half of the tube
[ə] ("uh")~500 Hz (mid)~1,500 Hz (mid)Uniform ~17.5 cm tube resonating at odd quarter-wave multiples

Because a uniform 17.5 cm closed-open pipe naturally resonates around 500 Hz, 1,500 Hz, and 2,500 Hz, leaving the vocal tract neutral produces a clear schwa [ə]. Every movement of the tongue perturbs those standing-wave pressure nodes and shifts F1 and F2 smoothly across the vowel space.

Why AudioWorklet matters for browser DSP

Neil Thapen’s 2017 release ran the Kelly-Lochbaum loop inside Web Audio’s legacy ScriptProcessorNode, which executed on the browser’s main UI thread alongside DOM layout and canvas drawing. On busy tabs or mobile phones, UI frames could delay audio buffers and cause clicks.

On pinktrombone.net, the waveguide runs inside a dedicated AudioWorkletProcessor thread, processing 128-sample blocks with smooth parameter interpolation even while the canvas animates at 60 fps and MIDI or Text-to-Speech events stream in.

Classic research behind Pink Trombone

Pink Trombone builds on decades of open acoustic-phonetics research originally cited by Neil Thapen in 2017:

  • Kelly, J. L. & Lochbaum, C. C. (1962) — Speech Synthesis, Proceedings of the Fourth International Congress on Acoustics (introduced the digital reflection-line waveguide).
  • Rosenberg, A. E. (1971) — Effect of Glottal Pulse Shape on the Quality of Natural Vowels, Journal of the Acoustical Society of America.
  • Lu, H. & Smith, J. O. (2000) — Joint Vocal Tract Synthesis with Nasal Coupling, CCRMA, Stanford University.
  • Story, B. H. (2005) — A Parametric Model of the Vocal Tract Area Function for Vowel and Consonant Simulation, JASA.
  • Mullen, J., Howard, D. M., & Murphy, D. T. (2006) — Waveguide Physical Modeling of Vocal Tract Acoustics, IEEE Transactions on Audio, Speech, and Language Processing.

Put the physics into practice

More Pink Trombone Guides