nx_audio_dred.nx

nx_audio_dred.nx -- DEEP REDUNDANCY for voice, the keystone for surviving bursty mobile packet loss WITHOUT a retransmit (at 150ms RTT a retransmit is a dead frame). The Opus-1.5 DRED idea, classical first: every packet carries a heavily-quantised "redundancy tail" = coarse copies of the previous K frames (just the LPC reflection envelope + energy, enough to resynthesise an intelligible frame). When a burst loses B<=K consecutive frames, the first packet that arrives after the burst carries coarse copies of the whole gap -> the receiver reconstructs every lost frame with no retransmit. K sets the recoverable burst length (~1s of audio at K=50 @ 20ms frames). Pure integer; the neural codebook (true DRED's 12-32kb/s for 1s) is the phone-CPU moonshot on top of this scheduler. license_tier: ORIGINAL