TT-FREIRAUM — User Manual

01The concept

A broadband ducker turns the volume down when the sidechain hits — the whole signal, including the parts that do not conflict. TT-FREIRAUM works spectrally: it analyses the sidechain every 10.7 ms (512-sample hop at 48 kHz, 11.6 ms at 44.1 kHz) and removes energy from the main signal only at the frequencies the sidechain actually occupies.

main out = main in  −  Σ  ( zone activity(t) × removed zone spectrum )

The removed energy is reconstructed per zone through its own overlap-add path and scaled by a per-sample time-domain activity — a hybrid of spectral weights (updated per hop) and analogue-style envelopes (updated per sample). A kick ducking a bass removes only the kick's bins; the bass's own harmonics in between keep playing. And because the removed part exists as a real signal, DELTA audition and ZONE SOLO come free, without phase errors.

02Quick start

1. Insert TT-FREIRAUM on the track that should make room (e.g. pads, guitars, a busy bus). The preset bar in the header offers factory starting points.
2. Feed the sidechain — either works, both are honest options: route the competing signal to channels 3/4 of this track (send destination 3/4, track channel count 4), or set SC SOURCE to GMEM and place a TT-TRUEAB PRE (free, same ReaPack) on the source track — cross-track ducking with no cabling at all.
3. Watch the header LED: it must read SC CH 3/4 SIGNAL or SC GMEM LINK OK. Then raise AMOUNT — the display shows the main spectrum, the sidechain spectrum and the live cut curve.
4. Press DELTA to hear exactly what is being removed, and trim the three zones to taste.

03Sidechain routing

The plug-in exposes four input pins — Main L/R on 1/2, SC L/R on 3/4 — and two ways to fill the sidechain pair:

For GMEM mode the sender needs both switches on: ACTIVE and DRY RING — the link carries the unprocessed audio itself, not just a level. Internally FREIRAUM then delays the main signal by 2048 samples before analysis, because the PRE may run before or after it inside the same audio block; the measured offset is tracked continuously, so the link stays sample-aligned. If the PRE stops updating for ~250 ms (bypassed, track offline), the link is declared stale and the plug-in falls back to silence on the sidechain: no ducking, clean passthrough. With several PREs running, the last writer wins.

04Zones & ballistics

The spectrum is divided into three zones by two crossover frequencies. The zone masks are complementary smooth-step crossfades in log-frequency, ±1/4 octave wide around each crossover, and sum to exactly 1 — the three zones partition the spectrum with no gaps and no overlap.

XOVER HIGH is constrained to at least twice XOVER LOW — raise the low crossover and the high one moves with it if needed.

Each zone has its own detector on the sidechain: the (pre-EQ'd, mono) sidechain is split by 4th-order Linkwitz-Riley-style band filters at the crossovers, followed by a fast peak envelope, a gate at THRESH with a 3 dB knee, and independent attack/release ballistics — computed per sample, so an attack can land well inside the 10.7 ms analysis hop (measured: 90 % after 2.9 ms on the LOW zone at its default ATK of 2 ms). The times are 99 % settling constants. Between hops the zone's spectral removal decays at its release rate instead of snapping back at frame boundaries.

Typical setting: fast low zone (kick vs. sub-bass, no phase smearing), a musical mid zone (vocal vs. guitars), a controlled high zone (hi-hat vs. snare crispness).

05Spectral engine

The engine runs a 4096-sample sqrt-Hann window on a 512-sample hop (8× overlap-add, constant-overlap-add sum 4). The sidechain is analysed with a 4096-point real FFT, which gives 2049 bins ≈ 11.7 Hz apart at 48 kHz; DC and Nyquist are never ducked. The main signal is zero-padded to an 8192-point transform, so the removal acts as a linear filter and its tails cannot wrap around inside the frame; the cut curve is interpolated onto the 5.9 Hz intermediate bins.

Two changes in v1.1.0 over v1.0.0: the hop is half as long (v1.0.0: 1024 samples, 4×), and the main-signal transform is truly zero-padded (v1.0.0 filtered circularly within 4096 samples). Measured with a sidechain tone that the main signal shares, at AMOUNT 100: the residue of a 400 Hz→2.8 kHz sweep drops from −55.4 to −64.2 dB, the residue of a 1 kHz vocal-style vibrato (±100 Hz, 6 Hz) from −46.7 to −51.2 dB. The energy of the removal's impulse response beyond 30 ms falls from −44.1 to −50.5 dB, removal away from the notch from −98 to −104 dB. Steady-state depth, attack/release, latency and the DELTA null are unchanged. The price is ~3.5 % of one core more.

Per hop, the sidechain magnitudes are converted to dB against a sine-peak calibration, mapped through THRESH/RANGE to a 0–1 weight, dilated by ±2 bins — a 4096-sample window cannot carve a notch narrower than roughly five bins, and without dilation a pure tone only reaches ~4.7 dB instead of the full 6.0 dB — and smoothed with a 1-2-1 kernel against comb artefacts and "watery" sound. The removal gain per bin and zone is min(1, AMOUNT × weight × trim) × zone mask; each zone's removed spectrum is synthesised into its own OLA ring, and the output subtracts the activity-weighted sum from the (latency-matched) dry signal.

06SC pre-EQ & gate

The SC PRE-EQ shapes the sidechain in the time domain, before both consumers — the spectral analysis and the zone detectors hear exactly the same filtered signal. One truth, no disagreement between "what is removed" and "when".

THRESH does double duty: it sets the bin level below which the spectrum counts as silent and the gate threshold of all three zone detectors (with a 3 dB knee). If a zone never opens even though the spectrum shows energy, THRESH is the first knob to lower.

07Controls & display

All sliders stay hidden by design — the drawn front panel (900×570) is the interface. Knobs: drag vertically, Shift for fine steps, mouse wheel over a knob, right-click or double-click for the default.

The spectral plot (log scale, 20 Hz–20 kHz) shows the main spectrum in grey, the sidechain in cyan and the current gain reduction as bars hanging from the top — green up to 12 dB, red beyond, 24 dB reaches half the plot height. The dashed THRESH line plus the RANGE band above it are shaded in, so you see exactly where the sidechain starts to bite. Below the plot, the TRIM strip holds a draggable handle per zone (±12 dB). The header shows live LAT … smp / … ms and GR … dB, the strongest per-bin reduction right now.

Options / Color

The OPTIONS button at the right end of the header bar opens a small box. Under COLOR it offers ten colour schemes for the whole front: ACID (the factory colour, a neon green), CYAN, ORANGE, MAGENTA, YELLOW, RED, BLUE, VIOLET, MINT and ICE. Click a swatch to switch the accent, its second colour and, for RED to ICE, the tinted panel background at once; a click anywhere outside the box closes it. The choice applies per plug-in instance, is stored with the project, the FX chain and REAPER presets, and does not touch any audio parameter. The factory colour is ACID, and the trace colours of the spectral display (MAIN grey, SC cyan, red for heavy gain reduction) keep their meaning in every scheme.

08Presets

The bar loads and saves the whole sound setting of the instance in one step (21 values): AMOUNT, the three zone TRIMs, both crossovers, attack and release of all three zones, the SC pre-EQ (HPF, LPF, the bell), LOOK, OUT, THRESH and RANGE. Not part of a preset: SC SOURCE (it describes your routing, not the sound), MIX, the audition modes (DELTA / SC LISTEN), the zone SOLO, LEARN together with its snapshot (a measurement of your sidechain, never stored), the colour scheme (OPTIONS) and the hidden test parameters. Loading a preset therefore never switches your routing, your wet/dry balance or an audition. Note that LOOK is part of a preset: a preset with lookahead changes the reported latency (REAPER re-aligns it automatically).

Factory presets

Starting points, not final settings. Each preset was measured with a pink-noise main signal (−18 dBFS) and a steady sine as sidechain (−6 dBFS), 3 s after the start, MIX 100 %, offline render. The table shows the level change of the main signal in a narrow band around the sidechain tone (±6 % for 1 kHz, ±10 % for 100 Hz), against the same setting with AMOUNT 0. In every case the level 250 Hz, 2.5 kHz and 6 kHz away from the tone stayed within ±0.01 dB.

The depth depends on the zone: a narrow sine in the sidechain is a small band, so the average over the measuring band is much smaller than the notch itself; with full-range sidechain material the carved gaps are wider. A factory preset changes exactly the 21 values above; SC SOURCE, MIX, audition, solo, LEARN and the colour stay as they are.

Where the user bank lives. Plug-ins cannot write files, so the 32 user slots are part of the plug-in state: they are saved with the project, with FX-chain and track-template copies, and in every REAPER preset of this plug-in (the + button in the FX window toolbar). To take your bank to another project or computer, save a REAPER preset once (+ → Save preset) and load it there: the whole bank comes with it. Projects saved with earlier versions load unchanged with an empty bank.

Typing names. REAPER keeps Space for the transport, so type Shift+Space for a blank. Depending on your REAPER settings Esc closes the whole FX window; use CANCEL instead. (The + menu of the FX window also has Send all keyboard input to plug-in.)

09Audition & solo

ZONE SOLO is a separate selector (OFF/LOW/MID/HIGH, one SOLO button per zone) that crossfades the output into a single zone's band: with NORM you hear that band after ducking — for judging whether a zone is being carved too hard; combined with DELTA you hear only what that zone removes. Measured zone isolation: −68 dB at 60 Hz and −120 dB at 9 kHz inside a MID solo.

All mode, solo and learn switches fade over ~20 ms — no clicks mid-track.

10Learn snapshot

Press LEARN while the sidechain plays: the running average of the sidechain spectrum (a ~200 ms window) is frozen into a static removal profile. Over a 20 ms ramp the weights crossfade from the live spectrum to the snapshot, and all three zone activities are pinned fully open — the result is an EQ-carving profile that keeps working even when the sidechain goes quiet, not a dynamic ducker. Release LEARN to ramp back to live tracking.

Let the sidechain play for at least a fraction of a second before pressing so the average is representative. A held profile survives transport stops while the instance runs — but it is not stored in the project or preset: after a reload the snapshot is empty, so a saved LEARN-on does nothing until you toggle it again with the sidechain playing.

11Latency & lookahead

LOOKAHEAD delays the main signal relative to the zone activity — the duck is already down when the (delayed) transient arrives, which kills the click at the onset (measured: onset leads by ~9–10 ms at LOOKAHEAD 10). At sample rates above ~93 kHz the lookahead tops out at 1856 samples instead of 20 ms (≈19 ms @96 k, ≈10 ms @192 k). The header shows the live LAT samples / ms.

12Hidden readouts

For diagnostics and test automation the plug-in publishes read-only parameters — hidden from the panel (JSFX - prefix) but exposed to the host, so they appear in REAPER's parameter/automation list:

13In practice

Three setups cover most of what TT-FREIRAUM is for. Each names the routing, the settings that matter and what to watch in the display.

Kick → bass: the classic duck, without cabling

Goal: every kick hit carves room in a sustained bass or pad — only where the kick actually lives, so the bass keeps its own mids and tops playing between hits.

1. On the kick track: insert TT-TRUEAB PRE (free, same ReaPack) first in the chain — it taps whatever plays at that point — and switch ACTIVE and DRY RING on. No sends to draw: the link runs over the shared ttab ring, across tracks and busses.
2. On the bass track: insert TT-FREIRAUM, set SC SOURCE to GMEM. The header must read SC GMEM LINK OK — NO PRE (GMEM) means the sender is bypassed, missing or has gone stale.
3. Raise AMOUNT to 50–80 %. Hold SC LISTEN and check that only the kick reaches the detector — if the channel carries rumble or cymbal bleed, raise PRE HPF until the listen contains just the body and the click.
4. LOW zone: ATK ≈ 2 ms, REL 20–60 ms — short enough to feel tight, long enough not to stutter. If the sub should dive deeper, reach for the zone TRIM before pushing AMOUNT further.

GMEM adds a fixed 2048-sample guard on top of the 4096-sample STFT latency — the header LAT shows the sum. On a rhythm-critical chain the channels-3/4 route avoids it (section 03).

Vocal unmasking: surgical carve on a bus

Goal: a dense bus (guitars, keys, pads) gives way to the vocal only inside the vocal band — everything outside roughly 300 Hz–4 kHz keeps playing untouched.

1. Send the vocal track to channels 3/4 of the bus (send destination 3/4, bus track channel count 4) and leave SC SOURCE on CH — the lowest-latency path, no link guard. Header: SC CH 3/4 SIGNAL.
2. Narrow the work band: XOVER LOW ≈ 300 Hz, XOVER HIGH ≈ 4 kHz. The MID zone is now exactly the band the vocal masks — the LOW and HIGH zones should stay near-silent and can even be trimmed down.
3. Voice the detector with the SC PRE-EQ: HPF ≈ 120–250 Hz against plosives and rumble, LPF ≈ 8–10 kHz so sibilance does not overdrive the cut, and a BELL boost (+4–8 dB, Q 1–2) on the masking region (~1–3 kHz) so precisely that band decides when to duck.
4. MID zone: ATK 5–10 ms, REL 200–400 ms — a slow release is what makes this transparent instead of pumping. AMOUNT 30–50 % is usually plenty for speech clarity.
5. Press SOLO on the MID zone while the vocal sings: you hear that band after ducking — if it sounds hollow, back the TRIM down. Then set SOLO back to OFF.

Before you commit: verify it is transparent

Goal: prove the duck removes exactly the collisions — and nothing else — then pin the behaviour.

• DELTA solos the removed energy: it should sound like the collisions, not the song — the slice of bass that collides with each kick hit, the guitar bins sitting under the vocal. If DELTA plays back whole program material, AMOUNT or the zone TRIMs are too hot; if it is near-silent while the display shows cuts, the sidechain path is the thing to check.
• SC LISTEN plays the sidechain after the pre-EQ — literally what the detector hears. Tune HPF/LPF/BELL here, not blind: if the listen contains the trigger and nothing else, the carve can only be surgical.
• ZONE SOLO is a diagnostic, not an effect — always return it to OFF. Its isolation is deep enough (−68 dB at 60 Hz inside a MID solo) that a forgotten SOLO reads as a broken mix.
• LEARN freezes the ~200 ms average sidechain spectrum into a static profile: the carve keeps working even while the sidechain pauses — right for "hold space for the lead through the whole section", wrong for punchy groove ducking. The snapshot is not stored in the project — re-learn after a reload.
• LOOKAHEAD 5–10 ms when onsets click: the cut is already down when the transient arrives. It is not free — check the cost in the header LAT readout (4096 smp base + lookahead, + 2048 in GMEM mode) and remember REAPER compensates all of it via PDC.

14Troubleshooting

• SC CH 3/4 NO SIGNAL — the send is missing, the track has only 2 channels, or the send lands on channels 1/2. Set the send's destination channels to 3/4 and the track channel count to 4.
• NO PRE (GMEM) — no TT-TRUEAB PRE alive: it needs ACTIVE + DRY RING on, and must actually run (a bypassed or offline PRE goes stale after ~250 ms). With several PREs the last writer wins — keep it to one.
• Ducking feels late on transients — raise LOOKAHEAD to 5–10 ms.
• Wide smearing instead of a notch — normal STFT physics; sharpen the result with a smaller RANGE and a harder THRESH, or narrow the sidechain with the pre-EQ bell.
• A zone never opens though the spectrum shows energy — THRESH gates the zone detectors too; lower it, or check the zone TRIM.
• XOVER HIGH refuses to go lower — it is held at ≥ 2 × XOVER LOW by design; lower the low crossover first.
• LEARN did nothing after reloading the project — the snapshot is not persisted; press LEARN again with the sidechain playing.
• Pumping on a busy sidechain — shorten the zone releases or drop MIX below 100 % for a subtler blend.