Audio Parallel Compression (New York Style)

Classic parallel ("New York") compression done the correct way: the dry signal stays at unity while an aggressive acompressor copy (threshold −25/−35/−45 dB, ratio 3/6/12, makeup +7/+13/+20 dB tuned so the wet path sits near unity at program level) is blended in at a wet mix percentage. The sum behaves as upward compression: quiet passages rise toward the loud ones while transients from the dry path survive — crest factor stays essentially unchanged while the quietest window of the file gains several dB. FFmpeg details verified on the installed build: acompressor threshold and makeup are linear amplitudes (not dB), attack/release are milliseconds, and detection=rms with link=average tracks the program envelope like a classic bus compressor; amix uses normalize=0 so the two paths add at true unity, with a limiter guarding the summed output. The tool measures the file in 8 equal windows before and after the render and reports the quiet-window lift, the loud-window lift, their difference (the upward advantage) and the crest factors, so you can see the effect instead of guessing by ear.