nx_voice_clone_synth.nx

nx_voice_clone_synth.nx -- VOICE-CLONE-001 rung 2: CREATE new audio in a captured voice. Given a voiceprint (the speaker's vocal-tract filter, as averaged PARCOR) this drives that filter with a glottal source at ANY requested pitch -> voiced speech in that speaker's voice, at that pitch. This is the "create voices" half of duplicate-and-create. Because pitch is the SOURCE and identity is the FILTER, the same voiceprint can read at a low pitch, a high pitch, or sing a melody (the singing-teacher tie-in) while staying recognizably the same speaker. Pipeline (no reinvention): PARCOR k (Q30, from nx_voiceprint) --step-up recursion--> LPC predictor a (Q30) ; glottal impulse train at fs/F0 as the residual --> nx_lpc_synth(residual, a) --> i16 PCM. license_tier: ORIGINAL module: nishi-core.audio.voice_clone_synth depends: nishi-core.voice.lpc_synth, nishi-core.audio.voiceprint capability: AUDIO_VOICE_SYNTH