nx_vox_tract.nx

nx_vox_tract.nx -- the TIMBRE half of animal-sound synthesis: a vocal tract as a physical tube. nx_vox_source produces a glottal flow. That flow sounds like a buzz until it passes through a throat, and the shape of the throat is most of what makes a rabbit sound like a rabbit and a dog sound like a dog. This organ is that throat, modelled as a chain of cylindrical sections. WHAT IS GENUINELY NEW HERE, AND WHAT IS DELIBERATELY NOT. A tree-wide audit ran before a line was written (operator: "make sure we arent duplicating stuff"), and it found that most of this problem is already solved: * nx_lpc_levinson ALREADY emits PARCOR reflection coefficients in Q30. Reflection coefficients are exactly what a tube chain is made of, so this organ adopts that representation verbatim rather than inventing a second one -- a tract built here is interchangeable with the voice stack's. * nx_lpc_synth ALREADY implements the all-pole filter in direct form. It is imported and USED. What did not exist anywhere in 16000+ files, and is therefore all this organ adds: * vt_area_to_k -- tube GEOMETRY (cross-sectional areas) -> reflection coefficients. The whole point: the existing coefficients are FITTED to a recording, these are DERIVED from a shape you can state, so a species can be described instead of sampled. * vt_k_to_a -- the step-up recursion, so a geometry can drive the existing direct-form filter. * vt_ladder -- the Kelly-Lochbaum scattering ladder. WHY A LADDER AT ALL, IF THE DIRECT FORM ALREADY EXISTS. They compute the same transfer function, so the ladder has to earn its place: it is the PHYSICAL structure. Each coefficient is one tube junction, forward and backward travelling waves are separately available, |k|<1 guarantees stability by construction, and a section can be moved while the filter runs (which is what articulation is). The direct form has none of that. To prove this is a second REALISATION and not a second IMPLEMENTATION, nx_vox_tract_gate asserts the ladder and nx_lpc_synth agree sample-for-sample on the same geometry. If they ever diverge, one of them is wrong and the gate says so. TRACT LENGTH SETS THE FORMANTS, and that is the species signature. For a uniform tube the resonances sit at (2n-1)*c/(4L), so a 150 mm dog throat puts its first formant near 570 Hz and a 15 mm mouse throat puts it near 5700 Hz. Section count is NOT free: a waveguide advances one section per sample, so N = 2*L*rate/c. The organ derives N rather than letting a caller pick a number that silently means a different tube than the one they asked for. license_tier: ORIGINAL expect_exit: 0