Live vocal de-noise you tune by ear.
Spill pulls drum and stage spill out of vocal mics in real time. A small neural network hears what's singer and what isn't, and you decide how far it goes: cut hard where the spill lives, and ease off wherever the vocal needs its air and body.
There's no lookahead, so it sits on a live vocal with about 2.4 ms of added delay.
More cut means a cleaner channel, but push too far and the vocal starts to lose its top end and its life. Spill gives you the controls to find that line for each singer, by ear, while the band plays.
One knob for how hard Spill cuts the spill, across every band. It's the first thing to set, and the curve works from it.
A low shelf, two bells and a high shelf, drawn over a live RTA. Pull a node down to cut harder where the kick and snare live; push it up to leave the vocal's body or air alone.
Spill knows when the singer is singing. Max GR while singing caps the cut so the vocal stays natural, and GR between phrases clears the channel when they stop. Threshold, attack and release set how it moves between the two.
The RTA shows the input, the output and what was removed, on top of each other. Delta listen plays only what Spill is removing: if you can hear the singer in it, back it off.
Set the key range to the part of the spectrum that's this singer, or compare against the other vocal mics, so a backing vocalist's bleed doesn't count as the lead singing.
Flip between two settings mid-song to compare them, undo any change, and save what works as a preset for the next show.
Spill doesn't gate the vocal, and it doesn't look ahead into the audio. The audio passes straight through a filterbank, and the model only ever adjusts band gains based on what it has already heard.
Nothing waits for future audio, so the only delay Spill adds is the filterbank's own.
Drop it on a vocal channel in any AU or VST3 host on your FOH Mac, Waves SuperRack included, mono or stereo. The editor is built around the RTA, so you set it by looking and listening.
Run every vocal mic in one app, with no plugin host needed. Each strip shows its own RTA and everyday controls side by side, with one master active / bypass switch and snapshot recall for the whole show.
A short wizard listens to the band without the singer, then the singer alone, for about 10 seconds each, and sets the threshold for that vocal mic.
If the licence is missing or expired, Spill passes the vocal through dry, latency-matched, exactly like bypass. It never cuts the vocal.
Your licence is only checked when Spill loads. Once it's running, nothing changes until you close it.
No account, no server and no internet connection needed. It works the same in a venue with no wifi.
Spill is in a small, closed beta. Each tester gets their own licence, so it isn't a public download yet.
It's free while it's in beta. In return I'd like to hear how it sounds on your shows and what breaks.