nx_voice_read.nx

nx_voice_read.nx -- VOICE-CLONE-001 rung 5-FUSE: a CLONED VOICE READING WORDS. Fuses the two halves of the arc: * rung 5a (nx_phoneme_synth) provides ARTICULATION -- WHAT is said (the phoneme sequence, the formant trajectory over time). * a parametric SPEAKER IDENTITY provides WHO says it -- a consistent vocal-tract-length warp (every formant scaled by warp_num/warp_den) plus the speaker's F0. Scaling all formants by one factor is VTLN (vocal-tract length normalization): it preserves the formant RATIOS that make /a/ an /a/, while shifting the whole spectrum to a different-sized tract -- i.e. a different person saying the same word. The phoneme inventory is DATA in formant-Hz (record = [F1,F2,F3,voiced,dur]); at read time each formant is warped to the speaker and converted to a 2-pole resonator coefficient a1 = 2*r*cos(2*pi*F/fs) (r=0.95, matching nx_phoneme_synth). NOTE (tech debt): fx.nx has a CORDIC cos, but it imports syscalls.nx -> nx_syscalls.nx, which collides with this arc's nx_syscalls_x86_64.nx (two files, same symbols, no dedup). Until the arc's syscall module is unified, an inline integer cos (Taylor + quadrant reduction, err ~1e-3, plenty for formant tuning) is used here. license_tier: ORIGINAL module: nishi-core.audio.voice_read depends: nishi-core.audio.phoneme_synth capability: AUDIO_SPEAKER_READER