nx_pitch.nx

nx_pitch.nx -- sovereign integer F0 (fundamental-frequency / pitch) detector via the YIN cumulative-mean-normalized difference function (CMNDF). Pure integer fixed-point, NO floats, NO syscalls, NO imports. This is the SHARED bottom rung of the Nishi audio arc -- one primitive that three threads all stand on: * voice clone -> a speaker's pitch range/contour is part of identity * singing teacher -> "am I on the right note?" = sung F0 vs target F0 * voice recording -> pitch stability / monotone detection (no-orphans "shared primitive" rule: build once, reuse everywhere.) Re-derived integer-only from de Cheveigne & Kawahara 2002 (YIN), patent-clean. Not CREPE, not pYIN, not any proprietary tracker. API (all buffers caller-owned; the organ touches no OS state): nx_pitch_f0(pcm,start,win,fs,min_lag,max_lag,out) -> F0_hz pcm: *u8 i16-LE mono PCM start: first sample index of the analysis window win: window length in samples (>= 2*max_lag recommended) fs: sample rate (Hz) min_lag: smallest lag searched (= fs / F0max) max_lag: largest lag searched (= fs / F0min) out: *i64 [0]=confidence Q14 (16384 - cmndf_min, clamped >=0) [1]=best lag (period, samples) returns: F0 = fs / best_lag (Hz), or 0 if degenerate nx_pitch_is_voiced(conf_q14) -> 1 voiced / 0 unvoiced license_tier: ORIGINAL module: nishi-core.audio.pitch capability: AUDIO_F0_PITCH