nx_voice_codec.nx

nx_voice_codec.nx -- NishiVoice v1: the composition primitive that orchestrates VAD + LPC + residual quantisation + FEC + frame into a single named sovereign voice codec. No Opus dependency. No external code. Pure composition of nxc2 substrate primitives. genealogy_id: nx_voice_frame_q10 + nx_lpc_q15 + nx_vad_q10 + nx_voice_fec_q10 + nx_voice_codec_tier_dispatch_q10 lineage_id: nishi_voice_codec_v1_q10 Composition pipeline (encode): PCM samples (Q-scale i64, frame-sized) -> nx_vad_classify -> verdict (SILENCE / VOICE / NOISE / ...) -> if SILENCE: payload = energy_byte (1 byte) + cn_seed (1) -> else: nx_lpc_analyze -> order-N coefficients in Q15 residual = 8 evenly-spaced subsamples (16-bit each) payload = [vad][order][coeffs * 2N][n_sub][sub * 2*n_sub] -> nx_voice_frame_build -> wire frame with hash-chain header -> nx_fec_encode_block appends FEC tail (half-res of previous frame) Decode reverses: frame bytes -> nx_voice_frame_parse -> payload bytes -> if vad=SILENCE: synthesise comfort noise samples -> else: read coeffs + subsamples; expand via piecewise linear interpolation v1 is the COMPOSITION skeleton. Bitrate optimisation (entropy coding of residual via nx_range_coder, codebook-VQ of LPC) is the v2 path documented in NISHI_COMMS_ROADMAP.md. The point of v1 is to prove every substrate primitive composes end-to-end with no external dependency.