Harshness Tamer

Works like the dynamic resonance suppressors used on brittle masters, built from FFmpeg primitives: the 2–5 kHz target band is band-passed and fed both into a sidechaincompress main input and its own sidechain detector, then recombined through an exactly complementary delta path — output = source + (compressed band − original band), computed with a negative-gain pan — so the band is swapped for its compressed self while everything outside it passes untouched, with no crossover phase cancellation. Before rendering, the band is measured with astats and the compressor threshold is placed a chosen headroom below that measured band RMS (gentle −9 dB, medium −6 dB, strong −3 dB), with ratios 2:1 / 3:1 / 5:1 and RMS detection — so steady content above the threshold gets pulled down by (level − threshold) × (1 − 1/ratio) while quiet passages pass unchanged. After rendering, the band is re-measured and the achieved reduction reported. Attack/release defaults (5/100 ms) are fast enough to catch cymbal swells without pumping the surrounding mix. Band edges are adjustable (1–6 kHz low, 1.5–8 kHz high) for pinning down a specific resonance.