ASSISTED COMP BETA

Controls

• LearnOff / On — Press On and play the track: it listens for 15 s of signal (or until you turn it Off, after at least 0.5 s), then stores what it heard and returns to Off.
• SourceVocal / Drums / Bass / Instrument / Mix — What is playing: Vocal, Drums, Bass, Instrument or Mix. Each has its own ratio.
• TargetNatural / Warm / Bright / Punchy / Radio / Podcast / Lo-Fi — The sound you want: Natural, Warm, Bright, Punchy, Radio, Podcast (also brings the loudness to -16 LUFS) or Lo-Fi.
• Amount0 to 150% — How much of the target's gain reduction is applied, 0-150%.
• Squash Trim−6 to +6 dB — More (+) or less (-) compression than the target: moves the threshold down or up.
• Speed Trim−2 to +2 — Faster (+) or slower (-): +1 halves attack and release, -1 doubles them.
• Mix0 to 100% — Blend between the dry and the compressed signal (parallel compression).
• Output−12 to +12 dB — Level after compression and makeup, on top of the automatic loudness of Podcast.
• LearnedNo / Yes — Hidden: whether Learn has run. Saved with the project.
• Learned Loud−80 to 0 dB — Hidden: the learned level of the loud parts (90th percentile of the 50 ms levels).
• Learned Median−80 to 0 dB — Hidden: the learned median level.
• Learned Crest0 to 30 dB — Hidden: the learned crest factor (peak over RMS).
• Learned Onset0 to 2000 ms — Hidden: the learned mean time between onsets (0 = none found).
• Learned Loudness−70 to 0 LUFS — Hidden: the learned integrated loudness (BS.1770).

Best uses

• A first compressor on a vocal without knowing the settings
• A podcast voice at -16 LUFS with one button
• Drum bus punch or parallel crush
• Gentle glue on a bus or a mix

Properties

Learn: level histogram of 50 ms windows (0.1 dB bins, only windows above -60 dBFS count), mean crest factor, onset intervals (a 5 ms envelope jumping 6 dB over a 100 ms one) and BS.1770 integrated loudness. Rules (tools/assisted_targets.json): the target's reduction on the loud parts times Amount sets the threshold; ratio per source plus a target step (1.5:1-10:1); attack 5-25 ms by crest factor times the target's factor; release 60% of the onset interval (50-500 ms); makeup gives back half the reduction; Podcast moves Output to -16 LUFS. Compressor: stereo-linked peak detector → 6 dB soft knee → attack/release smoothing → makeup → Mix → Output. No AI: fixed rules, the same result every time. Zero latency.

Formats

• JSFXJSFX for REAPER, with REAPER's own sliders
• GFXJSFX GFX for REAPER, with its own interface
• VST3VST3 for any DAW
• CLAPCLAP for any DAW
• LV2LV2 for Linux DAWs such as Ardour, Carla and Qtractor

Factory presets

Vocal Natural · Vocal Warm · Vocal Upfront · Radio Voice · Podcast Voice · Punchy Drums · Parallel Drums · Steady Bass · Instrument Glue · Mix Glue · Mix Punch · Gentle Leveler

Try it