This is the real two-input ducking chain, not a volume automation guess. Your music track and a separate trigger track (a kick, a vocal, a snare) feed FFmpeg’s sidechaincompress, which compresses the music only when the trigger exceeds threshold. The trigger path runs through an optional high-pass/low-pass pair first — high-pass the trigger to stop sub-bass rumble from mowing down your mix, low-pass it so cymbal wash cannot trigger the compressor; 0 disables either. Full envelope control: threshold in dBFS (converted to the filter’s linear scale), ratio 1:1-20:1, attack 0.01-2000 ms, release 0.01-9000 ms, makeup 0-24 dB, knee 1-8, sidechain gain ±36 dB, RMS or peak detection, average or maximum channel link, and a dry/wet mix for parallel ducking. Presets carry working numbers, not folklore: kick-duck bass 5:1 with 4 ms attack (fast enough to clear the kick, slow enough to avoid clicks on a bass) and 60 ms release; voice-over bed 2:1 at 30/250 ms with ~2-3 dB of dip; EDM pump with a deep threshold and 1 ms attack. After rendering, the tool decodes input, trigger and output and measures what actually happened: per-frame gain reduction while the trigger is active (12 dB above its floor), the baseline drift, active-frame share and a 24-point ducking curve. Note the envelope terms follow the compressor convention — attack is how fast gain reduction engages, release how fast it lets go.