AquaSynth/research/physical-voice-modeling/PEOPLE.md

Influential Physical Voice Modeling People

This note maps practical physical vocal-modeling researchers to AquaSynth language pressure. The point is not hero worship. It is to steal the useful organs cleanly.

Peter Birkholz

Primary practical artifact: VocalTractLab, especially the open backend API and 2.4 manual. The captured LPP and IRCAM talk pages frame VTL as a multi-level articulatory synthesizer with geometric tract modeling, gestures, vocal-fold models, acoustic simulation, and higher-level control.

Useful pressure for Aqua:

• Separate anatomy/morphology from control gestures. A vocal tract shape is not a bag of section knobs; it is a geometry that gestures deform.
• Keep multiple control levels: low-level physical controls for parity and differentiability, plus higher-level gestures and utterance controls that can drive those controls.
• Treat copy synthesis and optimization as expected workflows. The DSL should preserve named controls and bounded physical ranges so log-mel parity golf has meaningful axes instead of anonymous index soup.
• VTL is a reference implementation, not a runtime dependency. Aqua should capture the reusable semantics: continuous area curves, source ports, radiation ports, branch coupling, gestures, and articulatory targets.

Downloaded/captured sources:

• sources/implementations/vocaltractlab-backend-dev-main.zip
• sources/implementations/vocaltractlab-backend-readme.md
• sources/implementations/vocaltractlab-api.h
• sources/implementations/vocaltractlab-2.4-manual.pdf
• sources/papers/birkholz-2013-coarticulation.html
• sources/papers/gao-stone-birkholz-2019-copy-synthesis-ga.pdf
• sources/papers/promon-birkholz-xu-2013-training-continuous-acoustic-data.pdf
• sources/papers/weitz-steiner-birkholz-2017-gesture-tts.pdf
• talks/pages/peter-birkholz-vtl-2.3-lpp-talk.html
• talks/pages/peter-birkholz-ircam-physical-models.html

Brad H. Story

Primary practical artifact: TubeTalker and parametric vocal-tract area functions. Story's work is especially useful because it compresses tract shape into interpretable low-dimensional parameters while still grounding sound in airway acoustics.

Useful pressure for Aqua:

• Area functions are the right bridge between anatomy and waveguide lowering. Aqua controls should deform continuous length/area curves and let lowering choose emitted sections or fractional delays.
• Tube length and constriction shape should be continuous runtime parameters. The Mathur/Story/Rodriguez fractional-elongation paper is direct evidence that changing morphology should not require changing sample rate or recompiling a fixed section count.
• Resonance/formant behavior should be derived evidence, not the owner of the voice. Story's NCVS talk makes the source-filter teaching model legible, but the synthesis layer should own the tube and its standing-wave conditions.

Downloaded/captured sources:

• sources/papers/story-2011-tubetalker.pdf
• sources/papers/story-2005-parametric-area-function.pdf
• sources/papers/story-2013-phrase-level-airway-modulation.html
• sources/mathur-story-rodriguez-2006-fractional-elongation.html
• talks/pages/brad-story-ncvs-vocal-tract-resonances.html
• talks/transcripts/brad-story-ncvs-vocal-tract-resonances.txt
• talks/pages/brad-story-azpm-mufflers-voice-tracts.html

Julius O. Smith

Primary practical artifact: digital waveguide physical modeling, including the Stanford lineage that Faust's physical-modeling library clearly reflects.

Useful pressure for Aqua:

• Bidirectional delay-line wave variables are the efficient owner for 1D acoustic propagation. Scattering, reflection, loss, and radiation should be explicit graph semantics.
• Physical models earn their keep when they are cheap enough for realtime interaction. Faust-friendly lowering is not a compromise here; it is the performant version of the idea.
• Fractional delay and delay-line filtering are not afterthoughts. They are how continuous length, tuning, loss, and moving boundaries become audio-rate machinery.
• Passivity and bounded nonlinear junctions matter. Aqua controls should make explosive acoustic networks difficult to express accidentally.

Downloaded/captured sources:

• sources/faust-2.85.5-delays.lib
• sources/faust-2.85.5-physmodels.lib
• talks/pages/julius-smith-cirmmt-physical-modeling.html
• talks/transcripts/julius-smith-cirmmt-physical-modeling.txt

Sidney Fels and John E. Lloyd

Primary practical artifact: ArtiSynth, a 3D biomechanical simulation platform for dynamic vocal tract and upper-airway modeling.

Useful pressure for Aqua:

• Full biomechanical models prove that anatomy, tissue mechanics, contacts, and acoustic loading can be one integrated simulation. Aqua should not copy that weight into realtime Faust, but it should preserve the authority split: body geometry, mechanical gesture, airflow/acoustics, and rendered sound are distinct layers.
• 3D/finite-element truth models are good calibration and constraint sources for lower-dimensional waveguide renderers.
• Interactive timelines and logged model outputs are a reminder that Aqua's playground should expose physical controls and diagnostics, not only final sound.

Downloaded/captured sources:

• sources/papers/fels-2006-artisynth-vocal-tract.pdf
• talks/pages/sidney-fels-msr-artisynth-vocal-tract.html

Haskins, ASY, Gnuspeech, Praat, and SndKit

These are practical implementation pressure rather than a single-person lane.

Useful pressure for Aqua:

• Gnuspeech and ASY keep the older tube-resonance/articulatory lineage visible: source-filter language is useful, but the better owner is still an acoustic body with excitation.
• Praat's articulatory synthesis implementation is small and inspectable, useful as a sanity check for model records and tract object boundaries.
• SndKit's Pink Trombone material is a compact implementation reference for glottis and tract behavior. It should feed parity tests, not define the DSL.
• Animal acoustic communication keeps the syrinx/alien objective honest: source ports must be plural and topology-driven, not glottis-shaped forever.

Downloaded/captured sources:

• sources/implementations/gnuspeech-tube-resonance-model.pdf
• sources/papers/haskins-asy-animal-acoustic-communication.pdf
• sources/implementations/praat-articulatory-synthesis-manual.html
• sources/implementations/praat-vocaltract.cpp
• sources/implementations/sndkit-tract.html
• sources/implementations/sndkit-glottis.html