nx_aec_dtd_gate.nx

nx_aec_dtd_gate.nx -- proves the Geigel DOUBLE-TALK DETECTOR (aec_run_dtd) beats plain NLMS during double-talk. MEASURED head-to-head: same FAST step (mu~0.75, the one that diverged in nx_aec_gate test 2), same realistic signal (far-end talks ALONE first so the filter converges, THEN near-end joins = double-talk). The DTD freezes adaptation while the near-end speaks, so the converged filter is preserved; plain NLMS mis-adapts and lets echo leak through.