The only broadcast processor that is aware of the future
This is the sound of a station that means it. NG Processor pulls every record up to size, glues it with five bands, restores what the loudness war broke and lands it on a look-ahead true-peak final stage that flatly refuses to distort. Your listeners won't know why the station suddenly sounds expensive. You will.
Animated Future Timeline: the programme scrolls past a NOW line. Everything to the right of it is audio the Predictive engine has already heard, with its planned gain moves drawn on top.
Level it. Shape it. Protect it. A reference-class chain, measured against the hardware it replaces, and then, frankly, improved.
A floating-window rider with real upward room: long outros come up like the station is holding its breath with the listener, soft records land at full size, jingles are simply ignored, and a drum break is treated as a drum break: the gain holds through the stop and the music comes back without a dip. It has opinions. They are usually right.
Five bands that ride the record instead of squeezing it, common-mode calm against pumping and a shared bass budget that keeps the biggest kick of the night the biggest kick of the night. Pumping is something that happens to other stations.
A look-ahead limiter, an oversampled clipper and a crest shaper for real broadcast density, watched over by adaptive distortion guards. Separate sample-peak, true-peak and limiter ceilings mean one preset can feed a transmitter link and another a stream. An Output Trim after the last limiter sets the level for loggers and links without touching the density. The ceiling is not a suggestion.
Most AGCs either sit on their hands or panic when the next track slams in on top of a lifted fade. This one was built for exactly that moment, and it shows. An instant freeze ignores stings and fills, a fast assist catches the first beat of a real hand-off, and the slow rider lands the rest at its own pace. No dip, no pump, no apology.
Every processor ever built listens to what just happened and hopes for the best. Ours listens to what happens next. Five seconds of it. It knows the fade is a fade, the cut is a cut and the whisper is an intro, and its moves are planned to the sample before the audio gets anywhere near the processing. No hardware box does this. Hardware isn't allowed to know the future. We didn't ask for permission. A processor that knows how the song ends.
The engine knows the next track starts in 3.2 seconds, so it carries the fade up and drops the first beat exactly on the sample. No dip. No slam. Nothing smeared across the join. Somewhere, a competitor's AGC is still working out what just happened.
Brightness, density, attack and width are set for what is about to happen, not for what just did. A reacting chain breathes, pumps and apologises. Predictive simply arrives on time, every time, and never looks surprised.
The main window shows the programme ahead with the engine's decisions drawn on it: gain lift, transition, brighten. You see the move before you hear it. And when the programme does something the plan did not count on, the independent Shadow rider, an ordinary AGC that runs alongside all the time and never sees the plan, takes over and hands the level to the ordinary AGC within two seconds, also when a presenter arrives after the fade. The SHADOW GAIN meter shows you where it stands at any moment. Try that with a rack unit.
Classic is the real-time chain for a presenter on headphones. Predictive is for streams, automation, files and the NG Encoder plugin: anywhere five seconds of look-ahead cost nothing and the sound is all that counts. Same presets, same Easy Tune, one switch. Choose badly and it still sounds better than what you had.
Hand NG Processor a recording of a sound you want and the same programme before processing, and it builds a preset by doing what an engineer does: render the original through the real engine, measure the output against your recording, change a setting, measure again. It keeps only what brings it closer. Then it tells you, in numbers, how close it got.
Two recordings of the same programme: the processed sound, and the unprocessed source. For your own station that is your playout feed next to your old processor. For somebody else's station you will not get their feed; the closest you get is the records they played, from your own library, in their order. Jingles and talk are skipped. With only a processed recording you get an estimate, not a fit.
It matches the two recordings on content, then searches: density, levelling, tonal balance, stereo, fades, one family of settings at a time, scored at the output of the complete chain. A passage it did not train on decides whether the result stands. No guessing, no blending, no hidden magic.
A preset file, a report with reference, result and deviation for every measure, and a label: fitted, approximate or estimate. Your own old processor usually fits. Another make usually lands on approximate: the colour and the density come close, the way it breathes is its own. That is the truth, and the report shows it.
Replace the rack. Keep the sound. Lose the invoice. Record your old processor for an hour next to your feed and NG Processor learns your station. Other processors ship with a manual. This one takes requests.
We fed the same hour of programme through NG Processor and a flagship hardware broadcast processor with a price tag of around EUR 15K, and measured both outputs sample by sample against the untouched source. Same loudness, the same sonic signature, and cleaner peaks. We measured it against a 15K flagship. It brought its own report.
Clipped loudness-war masters get their peaks back, Revive hands them their dynamics back, hum and line whistles are notched the moment they are detected. All of it in front of the AGC, so the rest of the chain only ever sees clean, living audio. The loudness war is over. We are handling the clean-up.
A modern master arrives at the station already crushed: limited, clipped and flattened into a brick. Peak Restore rebuilds the tops that were sheared off. Revive goes further. Band by band it reads what the untouched source is doing and gives back the movement that mastering squeezed out, so snares crack, kicks land and the chorus lifts again. Then the chain processes a record that breathes instead of one that gasps. Three characters: Strong for the stations that want the punch back, Medium for most, Gentle for the ones that only want the breath. You will hear it on the first loudness-war hit of the day. Your listeners will just hear music.
An AGC fixes the level of a record, not its colour. The Band Leveler gives every band its own slow, patient regulator that draws each record towards the preset's target colour: dull records open up, boomy ones calm down, voices are left alone. Every record sounds like your station. None of them sounds like the others.
Flat-topped peaks are reconstructed, optionally with a spectral deep-restore pass, and a crest-gated dynamics stage returns transients to flattened masters within a bounded budget. Material that is already dynamic is left alone. We restore records; we don't remix them.
Parametric tone, a programme-aware presence lift and an air shelf with its own guard. Slow spectral normalisation corrects persistently dull or bright sources without touching artistic colour. The Balance voicing was measured against a reference-class hardware processor, and matches it. Without the rack, the fan noise or the invoice.
A flattened, clipped or brickwalled master gets less AGC, less five-band drive and less final drive, in proportion, judged by its densest loud section, held steady through verses and breakdowns. It only ever takes away; it never adds. Somebody in a mastering room already did enough.
A preset is a file next to the executable, not a rack. Eight of them carry a star, and each one is a finished station sound, voiced and measured, not a template with the knobs in the middle.
Classic is real time, for a presenter on headphones. Balance is the calibrated reference: neutral, measured against a reference-class hardware processor, the only one without Revive. Airwave is radio with muscle, the one most music stations will pick: outros lifted, first beats caught, highs polished. Pure restores instead of processes, no wideband AGC, no clipping, for people who own more cables than records. Coast is the big, wide sound without the five seconds.
Predictive hears five seconds ahead, for streams, automation and the encoder. Impact is forward and loud. Velvet is density without the edge. Vertigo is warm, big and unbothered. Horizon is the big commercial sound where every record lands on the same station.
Pick one and press TUNE: Easy Tune starts at zero around any of them, so the factory sound is the baseline and the slider is your signature. Behind it sits Expert, every parameter of the chain with its JSON name, its range and a plain-English explanation. Edit a factory preset and it simply becomes yours; the star goes, the sound stays. Seven of the eight ship with Revive on. All eight ship with an Output Trim of 0 dB and no apologies.
Input and output peaks, a CPU readout in the health bar, a bipolar AGC meter with net gain left of zero and attenuation right of it (plus, in Predictive, a dark-grey SHADOW GAIN row showing what the independent rider would do on its own), five band gain-reduction meters with peak hold, restoration and guard activity, and the final limiter on both channels. When the AGC bar stops walking left, you are looking at the standing-gain ceiling, not at a fault. Yes, we get that question. Now you have the answer.
The health bar watches live buffer pressure and clears itself when the cause is gone. It is a live indicator, not stale alarm history, and not a blinking LED you have to look up in a PDF.
The Input Balance Level tells the chain how loud your feed really is. A Calibrate button in Settings measures 20 seconds of your own programme and proposes the value, on the whole-dB grid, with everything that is wrong with the measurement (silence, clipping, a bad device format) refused rather than rounded.
64- or 4096-sample blocks render exactly the same output. Offline file processing equals the live path to the bit, and the same fade is judged the same way whatever moment the show started. Consistency is not a mood.
Identical channels produce identical output; stereo detection is linked so the image never wanders. Mono listeners, you're welcome.
Digital silence in, digital silence out: no denormal noise, no gain wound up in the gaps. Even our silence is clean.
The chain's latency is known exactly, so BYPASS stays in time (dry, with only the Input Level Correction applied) and file output stays aligned.
Filter corners and time constants are defined in hertz and seconds, not in samples, so the sound does not change with the rate. Your sample rate is your business. The sound is ours.
Dozens of behavioural tests must pass before a build exists. We would rather fail a test than your breakfast show.
Quick start, main window, settings, presets as files, Easy Tune, the AGC in depth, sound stages, final control, file processing, the NG Encoder plugin, troubleshooting and commissioning: all in the manual on this site. Read it once and you will know more about your processor than most people know about theirs.
2FM puts Hits, Retro, Late Night and The Edge on air with NG Processor, around the clock.
