nx_vox_source.nx

nx_vox_source.nx -- a SELF-OSCILLATING physical voice source. Rung R2 of the bioacoustics lane. WHAT MAKES THIS DIFFERENT FROM EVERY OTHER AUDIO ORGAN IN THE TREE. nx_audio_bark, nx_tts_formant, nx_phoneme_synth and friends are all SOURCE-FILTER synthesisers: something generates a pulse train at a frequency you tell it, and a filter bank shapes it. The frequency is an input. Nothing oscillates on its own, so nothing can do what real animal larynges do -- jump registers, split into two pitches, break into chaos -- unless a human scripts each event by hand. That is why nx_audio_bark's header says "tune by ear": there is nothing to measure because there is no physics. Here the frequency is an OUTPUT. This is a mass-spring-damper model of two vocal folds, driven by nothing but a constant subglottal pressure, and it oscillates because of the aerodynamics: * Each fold is two coupled masses (lower m1, upper m2) -- the standard two-mass idealisation of the body-cover structure of a real fold. * The glottal areas are a1 = g0 + xL1 + xR1 and a2 = g0 + xL2 + xR2. * Bernoulli gives the driving pressure P1 = Ps * (1 - (amin/a1)^2) where amin is the narrowest point. When the glottis is CONVERGENT (a2 < a1) the narrowest point is downstream, pressure inside the glottis stays high, and the folds are pushed APART. When it is DIVERGENT (a1 < a2) the flow separates at the entry, P1 collapses to zero, and the springs pull the folds shut. * Because m2 lags m1, the glottis is convergent while opening and divergent while closing. Energy goes in during opening and not during closing. That asymmetry IS the oscillation -- it is why a larynx works, and it is the whole reason this file exists. EVERYTHING THE RULER MEASURES FALLS OUT OF THE PHYSICS, unscripted: * f0 emerges from mass and stiffness; it is never written down anywhere. * SUBHARMONICS appear when the two folds entrain at 2:1 instead of 1:1. * BIPHONATION appears when left and right folds are detuned enough to stop entraining and each runs at its own rate. This is why the model is left/right asymmetric rather than a single fold mirrored -- a symmetric model can never produce it. * CHAOS appears at high subglottal pressure, where the collision nonlinearity dominates. * NON-REPETITION comes from turbulent pressure jitter, so no two cycles are ever identical. A dog habituates to a sample loop precisely because a loop cannot do this. The claim "these phenomena are emergent" is not an assertion in a comment -- nx_vox_source_gate SWEEPS asymmetry and pressure and shows the regimes appearing in the order the physics predicts, as measured by nx_bioacoustic_bench, which was built and frozen BEFORE this file existed. Integer fixed-point throughout (Q16), explicit Euler at an oversampled step. Deterministic: the same parameters give bit-identical audio, which is what makes a regression baseline meaningful. license_tier: ORIGINAL expect_exit: 0