Easy and hard voice onset in a noisy two mass model

Voice onset is a flow-induced instability: airflow through a nearly closed glottis makes the vocal folds oscillate on their own. I asked whether starting from a firmly closed glottis (hard onset) rather than a slightly open one (easy onset) changes whether oscillation starts, the subglottal pressure P_s it needs, or how fast it starts. I used a symmetric Ishizaka–Flanagan two-mass model with viscous flow losses (main model), 0.5 % pressure noise, and an adductory hold spring that kept the folds closed until it was released over 5, 20 or 100 ms. Onset was detected cycle by cycle from the flow ripple. Steady-pressure thresholds were the same for both starts: medians 200.9 Pa (easy) and 199.2–201.7 Pa (hard), with paired differences of at most one 2 Pa bisection step in most draws. This held for hold stiffness 500–5000 N/m, hold position −0.02 to −0.1 mm and onset cut-off 0.05–0.20, so H2 was rejected. Timing did differ. Hard onset began 0–5 ms after the release at full size, while easy onset grew from noise (median 62 ms at 1.05 × threshold, 12 ms at 1.2 ×). A 5 ms release started first near threshold; easy onset started first at 1.5 × or against a 100 ms release. This core pattern appeared in 76 of 100 wide random parameter draws, but where it switched depended on the cut-off and hold stiffness. Success rates did not differ (H3 not supported), onset delay fell with slope −1.6 rather than −1 (H4 partly supported), and no pressed no-onset region appeared (H1). Size scaling matched Lucero and Koenig; the stiffness–pressure boundary had the right shape but values about 2.4 times too high.