About Raèld Gate
Raèld Gate takes noise out of a recording and leaves the voice or instrument alone. It works in five parts, and most material needs only one or two of them:
- Spectral suppressor: follows the steady noise under the sound (room tone, hiss, fans, air conditioning, traffic) and turns it down frequency by frequency, only where the noise is.
- AI voice: a small neural network trained on speech. It protects the voice, or for speech it removes the noise itself, crowds and traffic included.
- Room: turns a room's late reverb down (de-reverb), keeping the sound's own presence.
- Hum & bleed: removes mains hum and buzz, and turns other instruments' spill down between a drum's hits.
- Gate: closes between phrases and hits, opening a moment early so no attack is lost.
The rule it is built on: take the noise down, never the voice or instrument. When the two can't be told apart (a whisper under loud noise), Raèld Gate keeps the sound and leaves some noise.
Installing
- Close your DAW.
- Unzip
Raeld-Gate-2.0.0-Windows.zipfrom the download page or your purchase email, runRaeld-Gate-Setup.exeand follow the steps. The plugin goes intoC:\Program Files\Common Files\VST3; choose "Plugin and standalone app" to install the standalone app too. - Open your DAW and rescan plugins. DAW plugin lists show it without the accent, as "Raeld Gate", so type "raeld" in the plugin browser.
Requirements: Windows 10 or 11 (64-bit) and any VST3 host, or the standalone app. If Raèld Gate 1.0 is installed, 2.0 replaces it.
Quick start
- Pick the SOURCE under the display: VOCAL, DIALOGUE, ACOUSTIC, GUITAR AMP, DRUMS or FULL MIX. It sets up the modules for that material. Or choose a preset close to your recording.
- Turn SUPPRESS (the big knob in the middle) until the noise between phrases is as low as you want. 50 % is a natural start; the arc of red lamps round it shows how far the noise can go down (0 to 40 dB).
- Press LISTEN: you now hear only what is being taken out. It should be noise. If you hear words, notes or attacks in it, bring SUPPRESS down.
- If the recording has an echoey room, raise ROOM.
The suppressor finds the noise on its own within the first second or so. If your recording starts with the voice straight away, or the noise is unusual, teach it: find a stretch where only the noise plays, press LEARN, play it, and press LEARN again. That noise print is saved with your session.
The main view
Raèld Gate's face is matte carbon black, and nothing on it is lit but red. The screen runs across the top with the gate meter and the IN and OUT meters beside it; SUPPRESS sits in the middle, a big flat disc of black anodised aluminium with a red slit of light and an arc of red lamps; the sources, LEARN and LISTEN are on the left, ROOM, OUTPUT and MIX on the right, and the module tabs and their panel below. Every knob is such a disc. The keys are in the head and the foot.
| Control | What it does |
|---|---|
| SUPPRESS | How far the noise goes down, 0 to 40 dB. Past the middle it also judges the noise more firmly, so the strongest settings clean deeper but can thin the quietest parts of a sound. |
| SOURCE | Six setups, one per kind of material. Choosing one sets the modules; SUPPRESS, ROOM, OUTPUT and MIX stay as they are. |
| LEARN | Press over noise alone, then press again: that noise becomes the print the suppressor starts from. HOLD LEARNED (SPECTRAL panel) keeps it fixed. |
| LISTEN | Hear only what is taken out (the delta). |
| ROOM | How far the room's late reverb goes down, up to 30 dB. For recordings that have a room: on a dry one it would shorten the natural endings. 0 = off. |
| OUTPUT | The level after everything, ±24 dB. |
| MIX | The cleaned sound against the original. Both are time-aligned, so MIX never comb-filters. |
| LATENCY | ZERO, LOW or HIGH (below). |
| BYPASS | The untouched input, time-aligned and crossfaded, so it never clicks or jumps. |
| MONO | Folds the output to mono, to check a mix in mono. Not saved in presets. |
| Presets, A / B, UNDO | The preset menu (factory and your own), SAVE, the two compare slots A and B with COPY between them, and UNDO / REDO (Ctrl+Z / Ctrl+Y). |
| SIZE | The window from 85 % to 300 % of the size it opens at (568 × 510 pixels); it keeps its shape and opens at the size you left it. |
The display. The upper half is the spectrum: the input (grey), the noise the suppressor has found (red) and the output (white). The lower half is the cut map: time runs left to right, frequency bottom to top, and the brighter the red, the more is being taken out there. Clean noise looks like an even red wash; the voice shows as dark shapes where nothing is cut.
The gate meter. The level the gate hears, and its cut from the top. The solid line is where it opens (AUTO shows the level it chose), the dashed line where it closes. Drag either line to set it (that turns AUTO off); double-click a line for AUTO again. IN and OUT beside it are the input and output levels.
SOURCE
| Source | What it sets up |
|---|---|
| VOCAL | Singing. The voice model protects the voice (HYBRID), and can never cut more than the classic suppressor. Gate off. |
| DIALOGUE | Speech, podcasts, voice-over. The voice model removes the noise itself (AI), crowds and traffic included; a gentle gate, keyed by the voice, closes the pauses. |
| ACOUSTIC | Acoustic instruments. The classic suppressor alone (the voice model is trained on speech); held notes and decays kept. |
| GUITAR AMP | Hum and buzz removed; a gate that ignores the low end closes between phrases. |
| DRUMS | A fast gate with look-ahead, and the bleed between hits turned down band by band. |
| FULL MIX | Mixes and live recordings. The classic suppressor alone, steady tones taken down gently; gate off. |
Latency
| Mode | Delay (44.1 / 48 kHz) | Use it for |
|---|---|---|
| ZERO | none | Tracking and live monitoring. The suppressor runs as 16 band filters, with the gate, hum and bleed; no voice model, no look-ahead. For light noise: its bands are too broad to clean heavy noise without touching the voice. |
| LOW | 512 samples (10.7 ms at 48 kHz) | The default. The full spectral suppressor and the voice model. |
| HIGH | 4096 samples (85 ms at 48 kHz) | Mixing and post: the finest frequency detail and look-ahead, and the bleed learned from a sidechain mic. |
The delay doubles at 88.2 / 96 kHz and again at 176.4 / 192 kHz (the same time in milliseconds). Raèld Gate tells your DAW its delay, so the track stays in time. Changing mode while playing fades out and back in.
GATE
The five tabs open each module's panel; a tab's light is on while that module works.
| Control | What it does |
|---|---|
| GATE ON | Turns the gate on. |
| AUTO | The threshold follows the noise floor. It learns the level of short spill (clicks, hi-hats) and opens only above it, but always under the quietest sustained sound louder than the spill. |
| ABOVE FLOOR | With AUTO: how far over the noise floor it closes (−6 to +24 dB). |
| THRESHOLD | With AUTO off: the level it opens at (−90 to 0 dB). |
| HYSTERESIS | How far under the threshold it closes, so it doesn't chatter (0 to 20 dB). |
| RANGE | How far it turns down when closed (0 to 80 dB). |
| RATIO, KNEE | An expander's slope under the threshold, up to a gate (1:1 to ∞), and how round the corner is. |
| ATTACK, HOLD, RELEASE | How fast it opens (0.01 to 100 ms), how long it stays open after the sound drops (to 2 s), how fast it closes (5 ms to 4 s). |
| LOOK-AHEAD | It hears up to 10 ms ahead, so it is open before a hit lands (LOW and HIGH). |
| ADAPTIVE | The release waits while a sound is still fading, so endings aren't chopped. |
| PEAK / RMS | Follow the key's peaks (fast, drums) or its average (smoother, voices). |
| KEY LOW CUT, KEY HIGH CUT | What the gate doesn't hear: below a frequency (a kick under a snare) or above one (hi-hats over a kick). |
| INPUT / SIDECHAIN / VOICE | What the gate listens to: the track, the sidechain input, or the cleaned voice (so it opens for the voice, not for the noise). |
| KEY LISTEN | Hear what the gate listens to. |
SPECTRAL
The suppressor itself; SUPPRESS and LEARN are on the main view. It follows the noise as it changes, holds still under held notes, and leaves a clean recording untouched. TONAL takes steady tones in the noise (a whine, hum lines) further down. HOLD LEARNED keeps a learned print fixed instead of following the noise from it.
AI VOICE
| Mode | What it does |
|---|---|
| CLASSIC | The spectral suppressor alone. |
| AI | The voice model removes the noise itself, crowds and traffic included, down to AI NOISE (up to 40 dB). For speech only: it was trained on speech and would cut singing and instruments. |
| HYBRID | The voice model protects what the suppressor would dent in a voice; it never cuts more than CLASSIC. Safe on anything. |
The voice model works in LOW and HIGH. On a clean recording it changes nothing.
ROOM
ROOM, on the main view, turns the late reverb down. Raèld Gate measures how long the room rings from the recording itself (the panel shows it, for example "ROOM FOUND: T60 0.47 s"). Held notes and the first 50 ms after each sound (its presence and tone) are never touched.
HUM & BLEED
| Control | What it does |
|---|---|
| HUM ON | Finds mains hum (AUTO: 50 or 60 Hz, or set it) and subtracts it, at no latency, following its drift. A bass note on the hum's frequency is kept. |
| HUM, BUZZ | How far the hum (harmonics 1 to 4) and the buzz (the higher harmonics, up to 4 kHz) go down, up to 60 dB each. |
| BLEED ON | Other instruments' spill between this track's hits turned down, in four bands; the hits untouched. |
| BLEED RANGE | How far the bleed goes down (0 to 40 dB). |
| SOURCE ON SIDECHAIN | Put the bleeding instrument's own mic on the sidechain (the kick's mic on a snare track). In HIGH the bleed is learned from it and taken down under the hits too. |
Presets
| Preset | For |
|---|---|
| Init | Vocal, SUPPRESS 50 %, the voice model protecting the voice, gate off. |
| Vocal Booth Clean | A good booth with a little hiss and air conditioning: a light touch that leaves the breaths. |
| Bedroom Vocal Rescue | A vocal in an untreated room with a computer running: the noise well down, the room shortened. |
| Podcast Voice | A talking voice cleaned of room tone and background; the pauses quiet but not dead. |
| Noisy Room Dialogue | Speech in a café, a street or an office: the voice model takes the crowd and traffic out. |
| Echoey Room Fix | A voice in a reverberant room: the late reverb down, the voice's own sound kept. Not for dry recordings. |
| Amp Hum Killer | A guitar amp's hum and buzz removed (50 or 60 Hz found by itself), a gate between phrases. |
| Snare Bleed Control | Hi-hat and kick between snare hits turned down; the hits untouched. |
| Kick Tighten | A kick's ring and spill cut short after each hit; the gate hears the low end only. |
| Live Stage Vocal | A stage vocal mic: the spill between lines turned down gently, the voice protected. |
| Acoustic Guitar Clean | Preamp hiss and room noise under an acoustic guitar; the decays kept. |
| Field Recording Rescue | Location sound with rumble, hum and machinery: everything steady down, the hum removed. |
Using the controls
- Drag a knob up and down; hold Shift for fine moves. The mouse wheel and the arrow keys work too. Double-click to reset. A readout shows the value while you adjust and when you hover.
- Every control has a tooltip.
- Everything can be automated in your DAW, and the controls follow.
Tips
- Use LISTEN often. If you hear words or notes in it, you are taking too much.
- Heavy noise: use LOW or HIGH, not ZERO.
- On speech, try DIALOGUE (AI). On singing and instruments stay with HYBRID or CLASSIC: the voice model was trained on speech.
- A crowd under a singer looks like wanted sound and is mostly left alone. If that is your case, LEARN on a gap.
- ROOM is for recordings that have a room. On a dry vocal it would shorten the singer's own endings.
- Parts of a sound that sit under the noise (a soft breath under loud room tone) go down with the noise: nothing can separate them.
What we measured
From our test set: synthetic singing with breaths, soft phrases and word endings under each kind of noise, drums with bleed, and real speech under real noise. Each case uses the preset made for it. "Noise" is how far it came down between phrases; "voice" is the voice's change (0 = untouched).
| Material | Preset (latency) | Noise | Voice |
|---|---|---|---|
| Singing, room tone 20 dB under | Init (HIGH) | −20 dB | −0.0 dB |
| Singing, room tone 10 dB under | Init (HIGH) | −20 dB | −0.3 dB |
| Singing, room tone 10 dB under | Init (ZERO) | −14 dB | −1.0 dB |
| Singing, fan 10 dB under | Init (HIGH) | −20 dB | −0.3 dB |
| Singing, computer hiss and whine | Bedroom Vocal Rescue (HIGH) | −26 dB | −0.0 dB |
| Singing, traffic | Init (HIGH) | −18 dB | −0.1 dB |
| Singing, 60 Hz hum and buzz | Init + HUM (HIGH) | −24 dB | kept |
| Singing in a room (T60 0.6 s) | Echoey Room Fix (HIGH) | late reverb −7 dB | −0.1 dB |
| Snare, hi-hat bleed 30 dB under | Snare Bleed Control (LOW) | −34 dB between hits | hits −0.0 dB |
| Snare, kick bleed, kick mic on the sidechain | Snare Bleed Control (HIGH) | −72 dB between hits | hits −0.0 dB |
| Speech, room tone 10 dB under | Podcast Voice (HIGH) | −27 dB | −0.5 dB |
| Speech, traffic 5 dB under | Noisy Room Dialogue (HIGH) | −30 dB | −1.0 dB |
- In LOW and HIGH, a recording with nothing to take out comes out bit for bit as it went in (delayed by the latency).
- CPU, stereo at 48 kHz with every module on: about 3.6 % of one core in ZERO, 4.3 % in LOW, 5.0 % in HIGH (measured on our development machine).
- Works at 44.1 to 192 kHz and any buffer size; mono, mono-to-stereo and stereo tracks, with an optional sidechain.
Coming from Raèld Gate 1.0
2.0 replaces 1.0: same plugin, free for 1.0 owners, same licence key. Your 1.0 sessions open with their settings translated (MODE becomes SOURCE, REDUCTION becomes SUPPRESS, the learned noise becomes the print, the gate keeps its settings), and your 1.0 user presets are rewritten once for 2.0 (the originals are kept beside them as .raeldpreset.v1). The sound is new, so a 1.0 session sounds close to before, not identical.
Mono and stereo
Raèld Gate runs on mono, mono-to-stereo and stereo tracks; your DAW picks automatically. On a mono-to-stereo track the mono source feeds both sides. A sidechain input is there for the gate's key and for drum bleed. The MONO button at the bottom of the window folds the output to mono, to check how the sound holds up on a mono speaker; it isn't saved in presets.
Activation and support
Paste your Raèld Gate license key, or your Raèld All Access key, into the activation window (LICENCE). A key works on 3 computers. Without a key the plugin runs as a demo with a short dropout every 45 seconds. See support, or email info@raeldaudio.com.
The AI voice model is RNNoise (Copyright Jean-Marc Valin, the Xiph.Org Foundation, Mozilla, Amazon and Mark Borgerding), used under the BSD-3-Clause licence; its full notice is installed with the plugin as Third-Party Notices.txt.
Raèld Gate is an original design and is not affiliated with or endorsed by any other manufacturer.