AquaSynth/research/vocal-source-grounding/README.md

Vocal Source Grounding

This packet grounds the next cut to Aqua's graph tissue-valve source law. The trigger was the Warbling White-eye audition: a low-drive, smooth single-labium candidate finally sounded birdlike, but the route there was suspicious. The question is whether low drive and harmonic suppression are legitimate source controls or compensators that should instead fall out of morphology, pressure, tissue loss, aperture geometry, and acoustic load.

Downloaded/captured source files live in sources/.

Decision Pressure

The current graph valve lowering still gives drive too much authority:

• it scales reservoir pressure, so it changes energy input;
• it participates in oscillation gating, so it changes phonation threshold;
• it scales modal/displacement contribution into flow, so it changes brightness;
• it sits beside pressure, damping, saturation, load_coupling, and rest_opening, which already have clearer physical meanings.

That is not one organ. That is a desk drawer full of half-labeled teeth.

Literature Anchors

Myoelastic-Aerodynamic Ownership

Elemans et al. 2015 provide the cleanest cross-species authority map: bird syrinxes and mammal larynxes both use the myoelastic-aerodynamic mechanism. Airflow supplies energy; tissue elasticity/restoring forces, tissue-wave phase, and aerodynamic asymmetry sustain oscillation. The paper explicitly ties frequency and mode of oscillation to mechanical tissue properties plus aerodynamic driving forces, not a standalone timbre-gain parameter.

Implementation implication: pressure or reservoir_pressure is the energy input. mass, stiffness, damping, layered phase, aperture geometry, and load decide what that energy becomes. A single drive knob should not also own spectral aggression.

Syrinx Models Use Pressure/Tension Paths, Not Timbre Rescues

Mindlin/Laje-style syrinx work models syllables as trajectories through physiological parameter space, especially pressure and stiffness/tension. Their review emphasizes that the syrinx itself can be a source of complex acoustical behavior, and that smooth excursions in pressure/stiffness parameter space can produce recognizable syllables. The same review highlights that source-tract and source-source coupling can create subharmonics, period-doubling, and complex periodic/aperiodic behavior.

Implementation implication: the higher-level controller can own motif gestures, but the lower-level source law must expose meaningful physical coordinates. Harmonic behavior should arise from nonlinear source dynamics and coupling regimes, not from a spectral smoothing knob.

Mechanical Syrinx Experiments Separate Bronchial Pressure And Tension

Elemans et al. 2009 modeled the syrinx as a collapsible-tube/starling-resistor style mechanical system with separately controlled bronchial pressure and membrane tension. The captured abstract reports high-frequency self-sustained oscillations, tension-dependent frequency, mass-dependent lower frequency limits, and strong coupling to distal tube resonance.

Implementation implication: source morphology needs separate controls for tension/stiffness, mass, damping/tissue loss, and tract load. Pressure should drive the system; distal tube/reflection should perturb it. A generic low-drive value is not the correct way to make a small bird source less harsh.

Two-Mass Vocal Fold Modeling Keeps Flow, Tissue, And Tract Separate

Ishizaka and Flanagan's classic two-mass model approximates the vocal folds as two stiffness-coupled masses, uses Bernoulli flow through the glottis, and represents the tract as a transmission line. It calculates glottal area, glottal volume velocity, mouth pressure, and relationships among subglottal pressure, cord tension, glottal area, and duty ratio.

Implementation implication: our layered valve law should move toward explicit upper/lower masses and aperture/flow equations. It should not hide duty ratio, open quotient, or source spectral tilt behind drive.

Self-Oscillating Fold Reviews Point To Coupled Nonlinear Physics

The 2024 review of synthetic self-oscillating vocal fold models frames phonation as coupled flow dynamics, tissue motion, and acoustics. It calls out turbulence, flow separation, large deformation/strain, collision, acoustic coupling, and radiated acoustics as difficult but central nonlinear phenomena. It also emphasizes that stiffness and geometry are meaningful experimental controls.

Implementation implication: if the current valve needs harmonic suppression, candidate owners are damping/tissue loss, collision stiffness/damping, aperture geometry, flow separation/loss, and radiation/load. A musical-synth style drive stage is the wrong default owner.

Vocal-Tract Loss Work Makes Area-Dependent Flow Loss A First-Class Actor

Birkholz and Haesner 2024 propose unified viscous and kinetic pressure losses in discrete tube vocal-system models, including the glottis. Their measurements support area-dependent viscous losses and partial pressure recovery at expansions; the losses affect formant bandwidths, airflow, and turbulent source power.

Implementation implication: some of the "harmonic suppression" we wanted by turning drive down should instead come from source/tract loss and radiation loading. For small apertures and narrow passages, area-dependent loss is not a post-EQ trick; it is pressure-flow physics.

Faust Can Express The Clean Version

Faust has first-class tools for this lane: recursive state with ~, filters, fractional delays, bidirectional waveguide primitives, mode filters, tube models, reed tables, and nonlinear functions. We do not need to keep a fake drive authority because of Faust. The hard part is choosing a source law that is simple enough to inspect and stable enough for realtime.

Implementation implication: the next cut can stay Faust-friendly if it is a small set of named locals: reservoir/downstream pressure, pressure drop, flow-resistance/loss, upper/lower displacement and velocity, aperture, collision/contact, load pressure, volume velocity, and radiation flow.

Authority Map For The Next Cut

Owner: AcousticSourcePort source morphology owns local valve dynamics; AcousticNetwork owns incident pressure/load and radiation; gesture/controller layers own time-varying physiological inputs.

Inputs:

• Controller/gesture: pressure, opening, tension, left/right activation.
• Source morphology: mass, upper_mass, lower_mass, stiffness, upper_stiffness, lower_stiffness, coupling_stiffness, damping, tissue_loss, collision_stiffness, collision_damping, rest_opening, aperture_shape, flow_loss.
• Graph load: incident/downstream pressure, tract impedance, tube losses, radiation opening/reflection/loss.

Outputs:

• Source volume velocity/flow injection.
• Optional debug locals for aperture, pressure drop, contact, tissue energy, load pressure, and radiated flow.

Derived state:

• Dominant pitch, harmonic richness, phonation threshold, breathiness, brightness, and biphonation are outcomes of the above. They are not owned by freq= or a generic drive.

Forbidden writers:

• No hidden fixed oscillator floor when stiffness exists.
• No direct displacement-as-audio lane except as a tiny diagnostic/legacy term.
• No "harmonic suppression" or "bird smoothness" scalar.
• No source-owned spectral EQ pretending to be morphology.

Proposed Cut, Not Yet Implemented

1. Split drive into physical pieces:

   • keep pressure/reservoir_pressure as the energy input;
   • introduce or repurpose flow_scale only as a unit/calibration factor;
   • move threshold/ease-of-oscillation into pressure drop, aperture, damping, and load;
   • move spectral harshness into aperture/contact/flow-loss/radiation.

2. Add explicit morphology fields only if they name physics:

   • tissue_loss or viscous_loss for dissipative tissue/flow losses;
   • aperture_shape for the nonlinear mapping from displacement to area;
   • flow_loss or flow_resistance for area-dependent pressure-flow loss;
   • keep collision_stiffness/collision_damping as the contact owner.

3. Change the source law shape:

   • compute pressure_drop = reservoir_pressure - downstream/load pressure;
   • compute aperture from rest opening + controlled opening + displacement;
   • compute flow from a pressure-flow relation damped by aperture-dependent resistance, not from arbitrary modal gain into tanh;
   • use modal/two-mass tissue state to modulate aperture and flow, not as a parallel oscillator output;
   • let tract/radiation losses shape brightness before any post-source normalization.

4. Keep a compatibility path only as explicit legacy:

   • drive can remain parsed for now, but it should lower to flow_scale or a deprecated alias with warnings/tests, not retain secret timbre authority.

Open Questions

• Do we want flow_scale exposed in DSL at all, or should it be a patch calibration constant hidden behind morphology presets?
• Should aperture_shape be a continuous scalar first, or should we jump directly to named aperture laws such as slot, wedge, membrane, and reed?
• How aggressive should the v1 area-dependent pressure-flow loss be before we have a calibration set for birds?
• Should one-mass law remain the default for realtime, with two-mass/body-cover reserved for richer source patches, or should syrinx defaults move directly to two-mass because MEAD phase asymmetry is central?

Source Index

• pmc-elemans-2015-universal-mechanisms.html and elemans-2015-universal-mechanisms.pdf
• pmc-new-perspectives-birdsong-physics.html
• wur-elemans-mechanical-model.html
• nokia-ishizaka-flanagan-two-mass.html
• pmc-self-oscillating-vocal-fold-review.html
• vocaltractlab-birkholz-2024-losses.pdf and .txt
• faust-physmodels-doc.html, faust-filters-doc.html, faust-noises-doc.html