Lozengrad
JSFX

Lozengrad Amfi is a complete guitar amplifier simulation plugin with a multi-stage signal chain: noise gate → pre-EQ → cascading soft-clip distortion → auto gain → mid EQ → cabinet simulator → post-EQ → Mid Spice → bus compressor → master volume. All EQ stages use RBJ biquads with pre-computed fixed coefficients.

Noise Gate (slider1, slider2): An envelope follower with 2 ms attack, 75 ms release, and 15 ms hold time. When the input level drops below the gate threshold (−40 to 0 dB), the gate closes smoothly, muting the input before the gain stage. Placed before Input Gain to prevent noise amplification.

Pre-EQ (fixed): A fixed analog-style shaping network before the distortion stage, designed to tighten the sound and prevent low-end muddiness. Consists of 4 cascaded HPF biquads at 95 Hz (Q=0.4), a peaking notch at 95 Hz (−4.2 dB, Q=0.11), a low-shelf at 108 Hz (−7 dB, Q=2.1), narrow peaks at 291 Hz (+1 dB, Q=0.75) and 120 Hz (+4.5 dB, Q=0.7), a wide presence peak at 2000 Hz (+13 dB, Q=0.2), and an LPF at 2026 Hz (Q=0.75).

Amp / Distortion (slider4): Three-stage cascading hard clipping with progressive gain staging. Signal is multiplied by 3 × drive, clipped to ±1, scaled by ×2.5, clipped again, scaled by ×2, and clipped a third time. This creates a saturation curve that transitions from clean to overdriven to fully distorted as Drive increases (0–24 dB).

Auto Gain: An envelope follower with a 90% smoothing factor normalizes the post-distortion level using auto_gain = 1 / (envelope + 0.1), then scales by 0.5 for a consistent output level across different drive settings.

Mid EQ (fixed): A fixed EQ placed between the distortion and cabinet stages, shaping the overall midrange character. Includes a 60 Hz HPF (Q=0.71), a low-shelf at 120 Hz (+6 dB, Q=1.3), peaks at 90.5 Hz (+6 dB, Q=0.6), 364 Hz (+1.3 dB, Q=0.6), 850 Hz (−3 dB, Q=0.35), 1320 Hz (+1.5 dB, Q=0.75), and 7400 Hz (+1.2 dB, Q=0.75). Scaled by 0.7079 after processing.

Cabinet Simulators (slider9): Three cabinet impulse response models built entirely from cascaded biquad filters. Cab1 models a modern high-gain 4×12" with 4 cascaded HPFs at 65 Hz (Q=0.7), a subtle low-end bump at 98 Hz (+1.5 dB), deep mid scoop at 492 Hz (−6 dB) and 552 Hz (−5.5 dB), presence peaks at 1515 Hz (+3 dB) and 2592 Hz (+1.3 dB), wide cuts at 2446 Hz (−3 dB, Q=6.7) and 3390 Hz (−1.6 dB, Q=5.2), and dual LPFs at 6000 Hz (Q=0.7). Output gain: +1.3335. Cab2 models a vintage open-back 2×12" with 3 HPFs at 107 Hz (Q=0.83), a low-shelf at 428 Hz (+6 dB), and broad peaks at 224 Hz (+1 dB, Q=1.4), 635 Hz (+5.4 dB, Q=2.2), 1658 Hz (+5.7 dB, Q=2.2), 3840 Hz (+5.7 dB, Q=1.8), and 8500 Hz (+6 dB, Q=3.0), with dual LPFs at 5350 Hz (Q=0.8). Output gain: 0.6683. Cab3 models a closed-back 1×12" with 2 HPFs at 75 Hz (Q=0.7), a complex midrange with peaks at 173 Hz (+2 dB), 245 Hz (+4.5 dB, Q=0.3 — sharp), 1864 Hz (+3.5 dB), 3200 Hz (+3 dB), and cuts at 354 Hz (−2 dB), 581 Hz (−6 dB, Q=5.8), 2800 Hz (−2.5 dB, Q=9.9), 6500 Hz (−5.8 dB, Q=11.9), and 11000 Hz (−3.8 dB, Q=11.0), plus dual LPFs at 5500 Hz (Q=0.7). Output gain: 0.7943.

Post-EQ (slider5, slider6, slider7): A 3-band tone stack after the cabinet: low-shelf at 100 Hz (Q=0.707), peaking at 500 Hz (Q=1.0), and high-shelf at 3200 Hz (Q=0.707), each with ±9 dB gain range. Coefficients are recomputed on every slider change for the selected Low/Mid/High values.

Mid Spice (slider10): A parallel saturation processor that applies a bandpass filter at 500 Hz (Q=0.8), drives the filtered signal with 0–24 dB of gain, applies soft clipping via the cubic Taylor approximation x − x³/6 (modeling the first term of a sine wave clipping characteristic), and blends the saturated mid band back into the dry signal. An output gain compensator 1 / (1 + (drive − 1) · 0.5) prevents level jump.

Bus Compressor (slider11, slider12, slider13): A feedback bus compressor at the end of the chain. Uses RMS detection (10 ms window) with a soft knee (6 dB), 4:1 ratio, 10 ms attack, and 100 ms release. Slider13 displays the metered gain reduction in dB (read-only). When disabled, the compressor is fully bypassed with unity gain.

Master Volume (slider8): Final output level with −24 to +24 dB range, scaled by 0.355 to prevent digital clipping through the full signal chain.

Lozengrad Amfi V2 is a ground-up rewrite of the original amplifier simulation with expanded architecture: a V3 noise gate with lookahead and hysteresis, selectable Classic/Hard amp modes, an interactive 3-band RBJ tone stack (Bass/Mid/Treble), power amp sag/presence/resonance simulation, bright cap, transient enhancer, and harmonic exciter. The signal chain has been restructured: V3 noise gate → input gain → pre-EQ → amp mode → bright cap → tone stack → presence → sag → resonance → mid EQ → cab sim → harmonic exciter → transient enhancer → DC blocker → master volume.

V3 Noise Gate (slider1, slider2): A significant upgrade from V1. Uses 2ms lookahead via a circular delay buffer (up to 512 samples) to detect transients before they arrive. The detection blends 70% peak (instant attack, 50ms release) with 30% RMS for combined transient and energy tracking. A hysteresis state machine (OPEN/CLOSED states with 6dB below threshold to close) prevents gate chatter at the threshold boundary. Dynamic hold time (20–80ms, scaling with signal level) keeps the gate open between transient bursts. When closed, a downward expander applies gain reduction with a 4:1 ratio instead of a hard mute — creating a natural fade rather than an abrupt cut. Separate attack (~6ms) and release (~100ms) gain coefficients ensure click-free transitions. The lookahead architecture means detection runs on current audio while the gated output reads from the delayed buffer, preventing transient loss at gate opening.

Input Gain (slider3): A fixed +21dB boost is baked into the gain formula: input_gain = 10^((slider3 + 21) / 20). At the default 0dB setting, the preamp receives +21dB to push pickups into the distortion sweet spot. The slider ranges from −24 to +24dB, providing an effective range of −3dB to +45dB. This is significantly hotter than V1's +6dB fixed boost, ensuring the distortion stages saturate properly even with low-output pickups in Hard mode.

Pre-EQ (fixed): Shares the same fixed analog-style shaping network as V1 — 4 cascaded HPF biquads at 95Hz (Q=0.4), a peaking notch at 95Hz (−4.2dB, Q=0.11), a low-shelf at 108Hz (−7dB, Q=2.1), narrow peaks at 291Hz (+1dB, Q=0.75) and 120Hz (+4.5dB, Q=0.7), a wide presence peak at 2000Hz (+13dB, Q=0.2), and an LPF at 2026Hz (Q=0.75). This pre-shapes the signal before the amp stage to tighten low-end and control pre-distortion fizz.

Amp Mode (slider4): Selects between Classic and Hard distortion architectures. Classic models a multi-stage tube amplifier with cascading soft-clip gain stages. Hard uses a simplified three-stage symmetrical hard clipper with auto gain normalization.

Classic Mode signal chain: Tube Stage 1 → Bright Cap → Tube Stage 2 + Cold Clipper → Tone Stack → Tube Stage 3 → Presence → Power Amp → Sag → Resonance. Tube Stage 1 applies (x · drive · 0.6) / (1 + |x · drive · 0.6|) scaled by 0.9. Tube Stage 2 uses a parallel asymmetric clipper: (x · drive · 0.8) / (1 + |x · drive · 0.8|) − 0.06 · (x · drive · 0.4) / (1 + |x · drive · 0.4|) scaled by 0.85 — blending fundamental and sub-harmonic saturation. The Cold Clipper adds +0.1 DC bias for asymmetric clipping: positive half-cycles use 0.7× gain, negative half-cycles use 1.2× (ratio ~0.58), modeling the cold-biased tube stage characteristic of vintage high-gain circuits. Tube Stage 3 is a fixed (x · 0.8) / (1 + |x · 0.8|) with 0.95 scaling. The Power Amp uses (x · 1.2) / (1 + |x · 1.2|). A soft limiter at the output clamps to ±0.99 via y = x / (1 + |x|). Final output: spl · master_vol · 3.0 · drive_comp.

Hard Mode simplifies the chain: Bright Cap → Hard Clipper → Tone Stack → Presence → Sag → Resonance. The clipper multiplies by 3 × drive, clips to ±1, scales by ×2.5, clips again, scales by ×2, and clips a third time — creating progressively saturated hard clipping. An auto gain envelope follower with 0.9 smoothing normalizes: auto_gain = 1 / (envelope + 0.1), scaled by 0.5. Final output: spl · master_vol · 0.355 — a fixed attenuation factor assuming auto gain provides level normalization.

Drive (slider5, 0–24dB): drive = 10^(slider5 / 20), ranging from 1× (unity) to ~15.85× (+24dB). In Classic mode, drive scales progressively into each tube stage. In Hard mode, it scales the first clipper stage at 3×.

Bright Cap (slider14): A 1-pole IIR HPF placed after the initial gain stage. Models the bright cap on a guitar amp's gain pot. Low mode applies a cutoff coefficient of 0.0002 (~3.2Hz at 48kHz), High mode uses 0.0006 (~9.6Hz) — subtle high-passing that preserves low-end body while letting upper harmonics bleed through. Implemented as y[n] = x[n] + bc_cutoff · (y[n−1] − x[n]).

Tone Stack (slider6, slider7, slider8): An interactive 3-band EQ replacing V1's post-cab Post-EQ. All three bands use RBJ biquad coefficients recomputed on every slider change. Bass is a low-shelf at 200Hz (Q=0.7, ±10dB via (value−5) × 4dB). Mid is a peaking filter at 800Hz (Q=0.8, ±10dB). Treble is a high-shelf at 3000Hz (Q=0.7, ±10dB). The series cascade creates the characteristic passive tone stack interaction where adjusting one band shifts the response of the others.

Presence (slider10, 0–100%): Models the negative feedback loop of a push-pull power amplifier. Implemented as an RBJ high-shelf at 4kHz (Q=0.7) with gain scaling from 0 to +8dB. At 50% the shelf provides +4dB boost above 4kHz; at 100% it reaches +8dB. The NFB topology shapes the power amp's response to high frequencies rather than adding EQ post-distortion.

Sag (slider9, 0–100%): Simulates power supply voltage sag under load. A peak envelope follower (2ms attack, 100ms release) tracks amplitude: sag_env = max(|L|, |R|) · sag_attack + sag_env · (1 − sag_attack). The target gain is 1 − sag_env · sag_amount · 2, clamped to 0.3 (−10.5dB). The gain is smoothed with the release coefficient before application. At high settings, sustained chords cause the virtual power supply to dip, producing the characteristic compression feel of a tube amp under load.

Resonance/Depth (slider11, 0–100%): An RBJ low-shelf at 100Hz (Q=0.9) simulating the power amp's resonance circuit. Gain scales from 0 to +8dB. Higher settings boost low-end before the cabinet sim, emulating the increased low-frequency coupling of a power amp driving a cabinet at higher volumes.

Mid EQ (fixed): Identical to V1 — a fixed post-distortion EQ: HPF at 60Hz (Q=0.71), low-shelf at 120Hz (+6dB, Q=1.3), peaks at 90.5Hz (+6dB, Q=0.6), 364Hz (+1.3dB, Q=0.6), 850Hz (−3dB, Q=0.35), 1320Hz (+1.5dB, Q=0.75), and 7400Hz (+1.2dB, Q=0.75), scaled by 0.7079.

Cabinet Simulator (slider13): Same three biquad-based cab models as V1 (Cab1/Cab2/Cab3) plus an Off option (slider value 0) that bypasses cabinet coloration entirely, allowing direct use with external IR loaders.

Harmonic Exciter (slider16, 0–100%): A parallel saturation processor that adds air and harmonics. An RBJ HPF at 5kHz (Q=0.707) isolates the high-frequency content, driven through soft saturation: y = (x · 2) / (1 + |x · 2|) · 0.5. The saturated high end is blended back into the dry signal proportionally to the amount control: output = dry + hex_hp · hex_amount. Adds shimmer and presence without harsh digital clipping.

Transient Enhancer (slider15, 0–100%): Dual-envelope attack boost. A fast envelope (3ms attack, 50ms release) and a slow envelope track the signal. The transient component is te_diff = max(0, te_fast − te_slow). Gain modulation: 1 + te_diff · te_amount · 3, so only material where the fast envelope leads the slow envelope (transient attacks) gets boosted — emphasizing pick attack while leaving sustain portions unaffected.

DC Blocker: First-order IIR HPF at ~0.16Hz: y[n] = x[n] − x[n−1] + 0.999 · y[n−1]. Removes DC offset from the asymmetric clipping stages.

Master Volume (slider12, −24 to +24dB): master_vol = 10^(slider12 / 20). In Classic mode, the output is spl · master_vol · 3.0 · drive_comp. In Hard mode, spl · master_vol · 0.355. The drive compensation factor drive_comp = drive > 1 ? 1 / (1 + (drive − 1) · 0.1) : 1 prevents excessive level buildup at high drive settings by attenuating proportionally to drive above unity.

Lozengrad Bas Amfi is a split-band bass amplifier simulation featuring a multi-stage preamp, power amplifier stage, transformer saturation, and an enhanced Speaker Coloration Engine. It splits the signal via an LR4 crossover, processes Low and High bands independently through compression, saturation, and multi-stage distortion, then passes through a power amp and a comprehensive cabinet simulator.

Signal Flow: Input Gain (×4 hidden boost) → Noise Gate → LR4 Crossover → Low Band (Mono Sum → Bass Focus peaking @ 80Hz → Compressor → Dynamic Saturation) + High Band (Pre-HPF @ 400Hz → Pre-Emphasis high-shelf → Pre-Dist EQ → Compressor → Frequency-Dependent Drive Modulation → Multi-Stage Distortion ×6 → Post-Dist EQ → Post-LPF → Tilt EQ) → Blend → RMS Auto-Gain → Power Amp (Comp → Low Damping HPF → Transformer Saturation → Soft Limiter) → Speaker Coloration Engine → Output Gain (×0.25 hidden cut) → DC Blocker.

Character Macro (slider12): A macro control interpolating across 14 DSP parameters with 5 breakpoints (Vintage/Warm/Punch/Aggressive/Modern). Controls preamp drive (2.5–14×), compressor threshold (−22 to −8dB) and ratio (3:1–10:1), saturation depth, transformer blend, cabinet resonance frequency and Q, low damping, presence boost, asymmetry, post-LPF cutoff (4–8kHz), pre-emphasis (6–22dB), clip blend, and dynamic response. One knob morphs the entire amplifier character.

Multi-Stage Preamp (slider7): A 6-stage cascaded distortion replacing the original single-stage clip. Stage 1 — gentle asymmetric tube preamp with dry blend. Stage 2 — polynomial soft clip (cubic Taylor approximation). Stage 3 — asymmetric DC bias for analog-style asymmetry. Stage 4 — moderate soft clip. Stage 5 — tanh glue stage. Stage 6 — asymmetry shaping with hard clip blend. Each stage is progressively saturated, creating a smooth, musical distortion curve that cleans up on lighter playing.

Frequency-Dependent Drive: A bandpass filter at ~1.2kHz (Q=1.5) measures mid-band energy and modulates the preamp drive depth in real time. When the mids are dominant (e.g., picked notes), the drive increases for harmonic richness. When the full spectrum is hot, the drive is naturally tamed — preventing fizz while maintaining grind.

Low Channel: The low band is mono-summed for low-end clarity. Bass Focus (slider5, 0–100%) provides a peaking filter at 80Hz (Q=0.5, up to +6dB). A character-driven compressor with soft knee and program-dependent release (3ms attack, 80ms release base) controls low-end dynamics. Dynamic low saturation (slider4, 0–100%) uses RMS/peak ratio to apply more saturation to sustain and less to transients, keeping the attack clean while adding harmonics to the tail.

High Channel: Pre-HPF at 400Hz strips low-mud before the distortion chain. Pre-emphasis high-shelf at 1.5kHz (6–22dB) ensures saturation energy. Pre-distortion EQ carves a deep scoop (100Hz −6dB, 350Hz −8dB) with an 800Hz+ shelf and 3.5kHz peak boost — both Attack-scaled (slider6). A character-driven compressor (0.5ms attack, 50ms release) tightens the high band before saturation. The 6-stage preamp follows, then post-distortion cleanup (mud cuts at 200Hz and 500Hz, de-hiss at 3kHz) and a character-dependent LPF (4–8kHz).

Tilt EQ (slider8): Opposition peaking filters at 60Hz and 3kHz (−6 to +6dB, Q=0.5). Negative = dark (boost lows, cut highs), positive = bright (cut lows, boost highs).

Power Amplifier Stage: Post-blend processing before the cabinet. Gentle compression (2:1 ratio, 6dB knee, 10ms attack, 100ms release) provides power amp sag. A character-dependent low-damping HPF (48–80Hz) tightens the low end. Transformer saturation blends iron (cubic nonlinearity) and nickel (asymmetric polynomial) cores with character-dependent weighting. A soft output limiter catches peaks.

Speaker Coloration Engine (slider10): A 10-stage biquad cabinet simulator with distinctive per-cabinet EQ signatures. Classic 8×10 — HPF 3×50Hz + LPF 2×5.5/5.3kHz, low-mid resonances (70Hz +3dB, 180Hz −1.5dB, 850Hz +1dB), mid scoop notch at 400Hz (−8dB), high-mid peaks/notches for cone breakup (2.5–5.6kHz), high-shelf presence, air shelf at 9kHz (−10dB), high-end saturation with speaker damping, 2 all-pass filters for minimum-phase emulation, and a character-dependent resonance peak. Modern 8×10 — wider bandwidth with extended low-end and high-end response, different resonance and scoop points. Dark 1×15 — lower cutoff with softer mids and reduced high end for a warm, vintage character. Each cabinet uses 20+ biquad filters designed to emulate the complex electro-acoustic response of real bass enclosures.

Noise Gate (slider2): Envelope follower with peak detection, 6dB hysteresis threshold, and smooth ramp (0.5ms time constant). Default −120dB = bypass.

Blend (slider9): Crossfades between dry and wet signals for parallel-style blending.

Auto Oversampling: Internal processing adapts to sample rate — 4× oversampling at ≤48kHz, 2× at ≤96kHz, bypassed at higher rates. Anti-alias filters protect the multi-stage distortion from foldback artifacts.

Lozengrad Bas Amfi Legacy is the original single-stage version of the bass amplifier, now maintained as a separate plugin. It uses the same split-band architecture with a simpler 3-stage hard clip distortion, basic cabinet simulation, and straightforward controls — for users who prefer the original voicing.

Signal Flow: Input Gain (×4 hidden boost) → Noise Gate → LR4 Crossover → Low Band (Bass Focus peaking @ 80Hz + Gentle tanh saturation) + High Band (Pre-HPF @ 400Hz → Pre-Emphasis high-shelf → Pre-Dist EQ → Asymmetric hard clip → Symmetric hard clip ×2 → Auto Gain → Post-LPF → Tilt EQ) → Blend → Cab Sim (3×HPF + 3×LPF + 4-band EQ) → Output Gain (×0.25 hidden cut) → DC Blocker.

Low Channel: Bass Focus peaking filter at 80Hz (Q=0.7, 0 to +6dB via slider5), followed by gentle tanh saturation (0–100%, slider4) with pre-gain (1× to 5×) and makeup compensation. Designed to thicken rather than distort the low end.

High Channel: Pre-HPF at 400Hz, pre-emphasis high-shelf at 1500Hz (+8 to +22dB via Character). Attack Pre-EQ (slider6) shapes response with 800Hz+ high-shelf and 3.5kHz peaking boost (0–8dB each). Three cascaded hard-clipping stages: asymmetric (+1 / −0.3 to −0.9) → symmetric ×2 → symmetric ×2. Post-LPF (3–6.5kHz) tames fizz. Drive auto-gain with 0.9-pole envelope smoothing.

Tilt EQ (slider8): Opposition peaking filters at 60Hz and 3kHz (−6 to +6dB, Q=0.5).

Character (slider12): Adjusts pre-emphasis (8–22dB), negative clip threshold (−0.3 to −0.9), hard clip blend, and post-LPF cutoff (3–6.5kHz). Lower = warm/round, higher = aggressive/grindy.

Cabinet Simulator (slider10): Classic 8×10 (HPF 3×50Hz + LPF 3×8kHz, low bump @70Hz +3dB, scoop @300Hz −8dB + @400Hz −10dB, presence @2.5kHz +4dB), Modern 8×10 (wider bandwidth, deeper scoop), Dark 1×15 (softer, darker voicing). All cabs use 36dB/oct slopes via triple cascaded biquads.

Noise Gate (slider2): Peak envelope follower with hysteresis, default −120dB = bypass.

Blend (slider9): Dry/wet crossfade for parallel processing.

Lozengrad Balans EQ is an 8-band parametric equalizer with HPF, LPF, Low Shelf, High Shelf, and Bell filter types per band, built-in FFT spectrum analysis, and a real-time combined EQ curve overlay. Each band uses standard RBJ biquad filters in Direct Form I topology with stability clamping, cascaded in series for the full 8-band response.

Band Architecture: Eight independent biquad filter instances per channel (16 total) process the signal in series cascade. Each band can be individually set to one of five types: HPF (high-pass, 12dB/oct), LPF (low-pass, 12dB/oct), Low Shelf (shelving with gain ±24dB), High Shelf (shelving with gain ±24dB), and Bell (peaking with gain ±24dB). The frequency range spans 20Hz to 20kHz with logarithmic slider mapping, Q ranges from 0.1 to 10, and gain from −24 to +24dB.

Parameter Management: The plugin uses a save-on-switch architecture — when you select a band (slider 1), the current slider values are saved to that band's persistent storage, then the new band's stored values are loaded into the sliders. This allows editing one band at a time with a compact slider layout while maintaining independent parameters for all eight bands. Frequency is computed from the normalized slider position as freq = 20 · 1000^(slider3) for logarithmic mapping.

Biquad Implementation: Each filter type uses the Robert Bristow-Johnson (RBJ) cookbook formulas with sin/cos precomputations. The HPF and LPF use a 12dB/oct structure (b0 = (1±cos)/2 family), while shelves and peaking use the full A-weighted formulas with sqrt(A) terms. All coefficients are normalized by a0 and clamped for stability: |a2| < 0.99999 and |a1| < 1 + a2. On coefficient changes, the filter state is attenuated by 99% to prevent zipper noise.

FFT Spectrum Analyzer: A 2048-point Hann-windowed FFT processes the mid signal (average of L+R) after EQ at sample rate. Magnitude bins are precomputed into pixel positions for the 640px-wide spectrum display, drawn as blue bars with 4-pixel spacing. The display updates at the FFT hop rate (every 2048 samples).

EQ Curve Overlay: The combined magnitude response of all 8 bands is computed on coefficient changes by evaluating each band's get_magnitude(f) function at 320 logarithmically spaced frequencies across the display. The product of all 8 band magnitudes is converted to dB and drawn as an orange/yellow curve over the spectrum analyzer, with ±18dB range and a center 0dB reference line.

Output Protection: A simple soft limiter clamps samples to ±0.99 at the output, providing basic protection against inter-sample overs from aggressive EQ boosts.

Lozengrad EQuinox is a simple tone-shaping EQ with adjustable low and high frequency bands, a built-in FFT spectrum analyzer, and a real-time EQ curve overlay. It provides straightforward bass and treble control with selectable frequency points, designed for quick tonal adjustments without the complexity of a full parametric EQ.

Low Band: A two-filter combination shapes the low end. A Low Shelf boost at the selected frequency (30–150 Hz) with Q=0.707 provides the primary low-end emphasis, followed by a Peak Filter cut at 1.5× the low frequency with Q=1.5. This pair creates a resonant low-end shape — the shelf lifts the sub-bass region while the peak notch attenuates the lower-mids just above it, resulting in a tight, focused low end with reduced muddiness. The Amount control (0–10) maps directly to both the boost and cut gain in dB.

High Band: Similarly, a two-filter pair shapes the highs. A Peak Filter boost at the selected frequency (3–10 kHz) with Q=1.0 provides the primary high-frequency emphasis, followed by a High Shelf cut at 1.2× the high frequency with Q=0.707. This creates a "presence peak" shape — the bell boost lifts the target frequency range while the shelf gently rolls off the air frequencies above it, preventing excessive sibilance or harshness. The Amount control works the same way, with positive values producing both the boost and the complementary cut.

Signal Flow: Input → Low Shelf Boost → Peak Cut (low corrective notch) → Peak Boost (high presence) → High Shelf Cut (air roll-off) → Safety limiter (±0.944) → Output. Each filter uses standard RBJ biquad formulas (low shelf, high shelf, and peak) with Direct Form I topology. The asymmetric filter pairing creates the characteristic "smile curve" shaping typical of analog tone stacks.

FFT Spectrum Analyzer: A 2048-point Hann-windowed FFT processes the mid signal after EQ, with per-pixel spectrum display across the full 640px width. Magnitude is normalized and displayed in dB from −72dB (bottom) to 0dB (top) with a 0dB center reference line. The analyzer updates every 2048 samples.

EQ Curve Overlay: The combined magnitude response of all four filters is computed in real-time by evaluating each filter's get_magnitude(f) at every horizontal pixel. The product of all four magnitudes is converted to dB and drawn as an orange curve over the spectrum analyzer, with ±15dB range and the 0dB center line for reference. This gives instant visual feedback of the total EQ contour.

Lozengrad Feeltre is a multi-mode filter with eight selectable filter types — LPF, HPF, BPF, APF, Tilt EQ, Low Shelf, High Shelf, and Tilt Shelf — with continuous frequency, Q, and gain control, plus a built-in FFT spectrum analyzer with real-time EQ curve overlay.

Filter Types: LPF (12dB/oct low-pass, standard RBJ biquad), HPF (12dB/oct high-pass), BPF (constant-skirt band-pass, gain peaks at center frequency), APF (all-pass, passes all frequencies with phase shift centered on the frequency). Tilt EQ uses a first-order shelving structure (b0 = G·g + 1, a0 = g + G with g = tan(w0/2)) that provides a smooth spectral tilt with ±12dB range — this is a one-pole filter with no Q control. Low Shelf and High Shelf are standard RBJ second-order shelves with Q. Tilt Shelf combines a Low Shelf boost at the selected frequency with a complementary High Shelf cut at the same frequency (same gain magnitude, opposite sign) using two filters in series — creates a steeper tilt than the first-order Tilt EQ.

Frequency Mapping: The frequency slider uses a custom log mapping with a power curve: freq = 20 · 1000^(slider2^0.8205). This provides finer resolution in the low-mid range where filter sweeps are most critical, while covering the full 20Hz–20kHz range. The mapping exponent 0.8205 creates a slight bias toward lower frequencies compared to a pure logarithmic scale.

Gain Behavior: For filter types 0–3 (LPF, HPF, BPF, APF), the Gain slider acts as post-gain — it's applied after the filter as a simple level adjustment. For types 4–7 (Tilt EQ, Low Shelf, High Shelf, Tilt Shelf), Gain is the actual filter gain — it shapes the magnitude response. In Tilt Shelf mode, the gain is split: +gain on the low shelf, −gain on the complementary high shelf, creating the tilt.

FFT Spectrum Analyzer: A 2048-point Hann-windowed FFT processes the mid signal after filtering, with per-pixel spectrum display from −72dB to 0dB, a 0dB center reference line, and blue magnitude bars at every pixel across the 640px width. Updates every 2048 samples.

EQ Curve Overlay: The combined magnitude response is computed per-pixel using each filter's get_magnitude(f) function. For Tilt Shelf mode, the product of both filter magnitudes is used. The resulting curve is drawn in orange over the spectrum analyzer with ±15dB range. The display also shows the current frequency value as an on-screen readout.

Output Protection: A safety limiter at ±0.944 prevents digital clipping. The filter state is reset whenever the filter type changes to prevent audible clicks from stale delay lines.

Lozengrad Kompresor V2 is a VCA-style compressor/limiter engineered for aggressive music production. Its advanced detector architecture, program-dependent release, dynamic saturation engine, and transient restoration make it equally capable of transparent bus compression and character-driven drum smashing.

Detector System: At the core of V2 is the Detector Blend (slider 8), which crossfades between peak and RMS detection paths. The RMS detector uses crest factor normalization to keep the threshold consistent regardless of blend ratio — so compression stays predictable across the entire blend range. Detector Thrust (slider 7) applies a 150Hz or 200Hz biquad high-pass filter to the detector signal, preventing kick and bass from over-triggering the compressor. This is critical in dense mixes where low-end energy would otherwise cause excessive gain reduction.

Dual-Stage Release: Unlike conventional single-time-constant releases, V2's release time dynamically adapts to gain reduction depth. When compression is deep, the release speeds up by up to 4x for quick transient recovery. Near unity gain, it returns to the user-set release time, minimizing pumping artifacts. The transition is continuous and seamless, computed as release_eff = release_fast * gr_depth + release * (1 - gr_depth).

Dynamic Saturation: As gain reduction increases, drive is automatically boosted through a unity-gain-normalized tanh function. At light compression the saturator is nearly transparent; at heavy GR it introduces rich harmonic density, emulating the natural behavior of an overdriven VCA chip. The normalization factor (tanh(x*drive) / drive) ensures output level stays consistent regardless of drive amount.

Punch Compensation (slider 9): Restores transient energy lost to compression by tracking the delta between fast-attack (0.5ms) and slow-attack (5ms) envelopes on the input signal. When punch is dialed in, this delta is added back to the processed signal, preserving the impact of drums and percussive elements even at high ratios and fast attack times.

Feedforward / Feedback (slider 10): Feedforward mode reads the detector from the input signal for aggressive, precise compression ideal for leveling and peak control. Feedback mode reads the detector from the compressed output, creating a smoother, more musical response with natural harmonics. Feedback also engages an additional envelope follower stage for improved gain tracking continuity.

Headroom Staging (slider 11): Low mode cuts 4.5dB at the input and compensates at the output, lowering the noise floor for cleaner operation. High mode pads the input and boosts the output, raising the noise floor but yielding a softer, more analog character. Normal mode passes through with no headroom adjustment.

Auto Makeup (slider 12): Instantly (sub-millisecond attack) tracks gain reduction and releases over ~300ms. This gives immediate response to transient GR changes while maintaining a stable average level over time. The makeup gain is computed from a smoothed average of the instantaneous GR.

Knee (slider 6): Three knee modes — Hard (0dB range, brick-wall transition), Medium (9dB gradual curve), and Soft (18dB smooth taper). The knee position is interpolated within the transition band using a quadratic curve, and the effective ratio ramps from 1:1 at the knee low boundary to the full ratio at the knee high boundary.

GFX Display: A scrolling waveform oscilloscope (green) and gain reduction histogram (red) using a 600-sample ring buffer updated at ~250Hz. The waveform includes reference grid lines at 0, -6, -12, -18, -24, and -36dB. A floating GR readout in the bottom-right corner shows the current gain reduction in dB with real-time smoothing.

Lozengrad Kompresor (the original, V1) is a VCA-style compressor/limiter with feedforward/feedback detection, a selectable high-pass sidechain filter (Thrust), dual-stage program-dependent release, a soft knee with three widths, and a subtle analog output stage saturation. It covers the full range from transparent bus compression to aggressive drum limiting.

Signal Flow: Input → Headroom Trim → RMS Envelope Detection → Thrust HPF → [Feedforward / Feedback] → Dual-Stage Release → Soft Knee Gain Computer → Gain Apply → Make-Up → Analog Saturation → Parallel Mix → Output.

Detector Thrust (slider6): An IIR biquad HPF placed in the detector path with three modes: Normal bypasses the filter (raw level detection); Medium applies a 150 Hz HPF (Q=0.707) to reduce bass trigger sensitivity; Loud applies a 200 Hz HPF (Q=0.5) for aggressive low-cut on the detector. This prevents kick drums and bass from over-triggering compression, allowing the compressor to respond more to midrange and transient content.

Tone Type (slider7): Feedforward mode applies the derived gain reduction directly to the input — clean and precise, with immediate transient response. Feedback mode feeds the compressed output back into the detector, creating a smoother, more musical compression with natural harmonic emphasis, characteristic of analog VCA designs.

Dual-Stage Release: The release time is program-dependent: when the compressor is deep in gain reduction (gain near minimum), a fast release coefficient (4× faster than the user setting, minimum 5 ms) takes over for quick transient recovery. As gain approaches unity, the release blends back to the user-specified time. This prevents pumping on fast transients like drums while maintaining smooth decay on sustained material.

Soft Knee (slider5): Three knee widths: Hard (0 dB) — brickwall transition at threshold; Medium (9 dB) — gradual onset over a 9 dB window; Soft (18 dB) — very smooth, subtle compression onset over an 18 dB window. The knee is computed via linear interpolation of the compression ratio over the knee window.

Headroom (slider8): Scales the internal signal level before and after compression: Low (−4.5 dB input, +4.5 dB output) for conservative headroom; Normal (unity); High (+4.5 dB input, −4.5 dB output) to push the detector harder for more aggressive response.

Output Saturation: A fixed analog-style tanh soft-clipper at the output stage with 1.08× drive (~0.7 dB). Implemented as tanh(x · 1.08) / tanh(1.08), it adds barely-perceptible even/odd harmonics, only coloring signal peaks while remaining transparent at normal levels. The denominator normalizes the output to match the unity gain point of the saturation curve.

Parallel Mix (slider10): Allows blending 0–100% of the compressed signal with the dry input, enabling New York-style parallel compression without external routing.

GFX Display: A 32-pixel-tall gain reduction meter showing a horizontal bar graph with dB grid lines every 3 dB (0 to −24 dB). The current GR value is displayed numerically in the top-left corner. The meter uses exponential smoothing with a 300 ms decay constant for a clean, readable display.

Lozengrad MB Saturasyon V2 is a complete rewrite of the original multiband saturator, featuring real oversampling, LR4 Linkwitz-Riley crossovers, four physically-inspired saturation models per band, a Character macro, and a hybrid auto gain system. It processes Low, Mid, and High bands independently with per-band mode, drive, and character controls, then sums them through a selectable output stage.

Crossover Network: Three bands are split using cascaded biquad filters in an LR4 (Linkwitz-Riley 4th-order, 24dB/oct) configuration — two biquad stages per crossover point per channel. The Low/Mid crossover is adjustable from 50–500 Hz, and the Mid/High from 800–8000 Hz. Filter coefficients are recalculated on slider changes, and all delay lines are flushed on crossover or mode changes to prevent clicks. The crossover operates at the oversampled rate when oversampling is active.

Saturation Modes (sliders 3, 7, 11): Each band independently selects one of four modes. Tape uses a piecewise transfer function with threshold/ratio saturation (threshold decreases and ratio increases with Character), followed by tanh normalization — emulates analog tape compression and saturation. Tube uses asymmetric exponential curves with DC bias and negative envelope feedback — creates even-order harmonics with a warm, asymmetric vacuum tube character. Soft applies a knee-based polynomial below threshold and tanh beyond — clean at low levels, rounding off at higher ones. Hard uses tanh with a limit ceiling that narrows as Character increases, plus an optional post-boost at high Character — aggressive, in-your-face clipping. All saturators accept a drive (0–24 dB) and Character (0–100%) parameter per band.

Character Macro (sliders 2, 6, 10): A continuous control that morphs the saturation character from clean (0%) through vintage (50%) to aggressive (100%). Each saturation mode interprets Character differently: Tape adjusts the knee threshold and ratio, Tube increases asymmetry and bias, Soft narrows the linear region, Hard reduces the ceiling limit. This gives a single macro knob for tonal shaping without needing to change modes.

Hybrid Auto Gain (sliders 4, 8, 12): Per-band automatic makeup gain that blends short-term RMS and peak tracking. An envelope follower measures the input RMS level of the band, compares it to the output RMS of the saturated band, and computes a target gain. The gain is smoothed with a one-pole filter (0.999 coefficient, ~1000-sample time constant). The target is clamped between -12dB and +12dB (factor 0.25–4.0). This ensures perceived loudness stays consistent regardless of drive setting.

Oversampling (slider 18): Selectable 1x, 2x, 4x, or AUTO. In AUTO mode, the oversampling factor is determined by the highest saturation mode index: Tape/Tube (mode 0–1) use 2x, Soft/Hard (mode 2–3) use 4x. When oversampling is active, the input is linearly interpolated between samples (fractional delay based on oversample index), processed through the crossover and saturation at the higher rate, then decimated by averaging all oversampled results back to the original rate. This significantly reduces aliasing from the nonlinear saturation functions.

Output Stage (slider 19): Four post-saturation bus processors. Transparent passes the summed bands through unchanged. Soft Clip applies a tanh function with a 0.95 ceiling for a final gentle rounding. Transformer uses a cubic nonlinearity x + 0.03 · x · |x| clamped to ±1.0 — simulates core saturation of an audio transformer. Tape applies a final tanh at 0.85 ceiling — emulates the global saturation of analog tape across the full signal.

Listen & Delta Modes (sliders 20, 21): Listen mode allows soloing individual bands (All / Solo Low / Solo Mid / Solo High) for targeted EQ and gain staging. Delta mode outputs the difference between the processed signal and the dry signal — useful for hearing exactly what the saturation adds, or for parallel processing with external mix control. The dry path uses a 4-sample delay line to phase-align with the wet path's crossover filter latency.

Signal Flow: Input Gain → Linear interpolation (if oversampling) → LR4 crossover (LP1 → LP2 for Low, HP1 → LP2 for Mid, HP1 → HP2 for High) → Per-band saturation (drive + character + mode) → Auto gain → Band sum → Listen solo → Delta subtract → Output stage → Dry/Wet mix → Output Volume. All bands share identical processing structure with independent parameters.

Lozengrad MB Saturasyon (the original, V1) is a 3-band multiband saturator that splits the signal with Linkwitz-Riley 4th-order (24 dB/oct) crossovers, applies one of four saturation modes to all bands with independent drive and gain control, and recombines them. It features selectable oversampling (2× or 4×), per-band RMS auto gain, solo listen modes, and dry/wet mix.

Crossover Network: Three bands are split using cascaded biquad filters in an LR4 configuration — two biquad stages per crossover point per channel. The Low/Mid crossover sweeps 50–500 Hz and the Mid/High crossover sweeps 800–8000 Hz. Coefficients are recalculated at the oversampled rate when either crossover frequency or OS mode changes.

Saturation Modes (slider15): A single mode applies to all three bands. Soft Clip uses x / (1 + |x|) — an algebraic soft clipper with smooth rounding at the threshold. Tube uses an asymmetric exponential: x > 0 ? 1 − exp(−x) : −1 + exp(x) — creates even-order harmonics with a warm vacuum tube character. Tape uses (2/π) · atan(x) — arctangent saturation modeling analog tape's gradual compression-like saturation curve. Hard Clip applies a brickwall max(−1, min(1, x)) clamp for aggressive square-wave clipping.

Per-Band Controls: Each band has independent Drive (0–24 dB, slider4/7/10) and Band Gain (−24 to +24 dB, slider5/8/11), plus a per-band Auto Gain toggle (slider6/9/12). The Auto Gain system uses an RMS envelope follower comparing input and output RMS levels per band, smoothed with a 0.999 pole, and normalizes the saturated output to match the input level — compensating for the volume change introduced by saturation while preserving the harmonic character.

Oversampling (slider17): Selectable Off, 2×, or 4×. The entire processing chain (crossover, saturation, auto gain) runs inside an os_factor loop, and the accumulated samples are averaged down: wet = sum / os_factor. Unlike V2's multi-rate polyphase cascade, V1 uses a simple uniform oversampling approach.

Listen Mode (slider16): Solo All Bands, Solo Low, Solo Mid, or Solo High — useful for dialing in individual band settings without removing the crossover interaction.

Signal Flow: Input Gain → LR4 Crossover (Low / Mid / High) → [Saturation + Auto Gain + Band Gain per band] → Sum → Listen Solo → Output Gain → Dry/Wet Mix. Input Gain spans −24 to +24 dB, Output Volume spans −24 to +24 dB, and Mix ranges from 0–100%.

Lozengrad Klip is a precision clipper/limiter built for transparent peak control and aggressive saturation alike. Four distinct clipping modes, selectable oversampling, harmonic tilt shaping, and frequency-weighted clipping make it suitable for everything from mastering ceiling control to drum bus destruction.

Clipping Modes (slider 4): Hard Clip applies a brick-wall max(-1, min(1, x)) clamp — pure square-wave hard clipping. Soft Clip uses a cubic polynomial x - x³/3 for a rounded knee with natural saturation harmonics. Tube Clip implements x / (1 + |x|), an algebraic approximation of vacuum tube saturation with smooth asymptotic clipping. Tape Clip uses (2/π) · atan(x · π/2) to emulate analog tape's soft saturation curve with a gradual, musical compression-like feel.

Harmonic Tilt (slider 6): Shapes the even-order harmonic content of the clipped signal by applying a cubic nonlinearity and DC bias before the clipping stage. At low values, the tilt is minimal and the clipper stays clean. At higher values, asymmetric saturation introduces rich even-order harmonics, adding warmth and analog character. The output is normalized to prevent level buildup from the nonlinear pre-processing.

Frequency Weight (slider 7): Implements a pre-emphasis / de-emphasis filter around the clipping stage. A first-order difference equation pre = in - α · prev_in boosts high frequencies before clipping and post = out + α · prev_out restores them after. This makes high-frequency clipping artifacts less audible by shaping the noise floor — the clipper works harder on pre-emphasized highs while the de-emphasis masks foldback distortion. The alpha coefficient scales linearly up to 0.874.

Oversampling (slider 5): Selectable 1x, 2x, or 4x oversampling using a 23-tap symmetric half-band FIR interpolator/decimator. The half-band filter coefficients provide a flat passband up to ~0.45π with >60dB stopband attenuation. At 2x and 4x modes, the zero-stuffed samples pass through the HBF structure before and after clipping to reduce aliasing. PDC (plugin delay compensation) is reported to the host: 11 samples at 2x, 16 samples at 4x.

GFX Display: Scrolling waveform oscilloscope (600-sample ring buffer at ~250Hz update rate) with a horizontal threshold line. Signal below threshold draws in green; signal exceeding the threshold draws in red above the threshold line, giving instant visual feedback on how much material is being clipped. Reference gridlines at 0, -6, -12, -18, -24, -36dB for level orientation.

Signal Flow: Input Gain → Pre-emphasis (Frequency Weight) → Threshold scaling → Clipping (with Harmonic Tilt) → Output scaling → De-emphasis → Dry/Wet Mix → Output Gain. The threshold operates as a gain-scaling stage before clipping: the signal is divided by the threshold, clipped, then multiplied back, making threshold act as a drive control relative to the hard ceiling.

Lozengrad MicroDelay is a procedural micro-delay modulation engine for stereo width and depth enhancement. Instead of conventional LFO-based chorus or delay, it uses a discretized Mackey-Glass delay differential equation (DDE) to generate deterministic chaotic modulation — producing organic, non-repeating stereo expansion that avoids the periodic "seasickness" of traditional LFO modulation.

Core Architecture: The signal is encoded into Mid/Side (M = L + R, S = L - R). The Side channel is fed through a variable-length delay line modulated by the Mackey-Glass system, then blended with the original Side before Mid/Side decoding. This means the delay modulation only affects the stereo difference signal, preserving mono compatibility while creating rich, evolving spatial depth.

Mackey-Glass Modulation Engine: A chaotic DDE originally developed to model physiological dynamics, adapted here for audio modulation. The equation x[n] = x[n-d] · β / ((1 + x[n-d]^γ) · (1 + α)) is solved with γ = 10, producing deterministic non-periodic behavior. Mod Rate maps to the MG parameters α (0.08–0.30), β (0.20–0.80), and delay d (10–30 samples), controlling chaotic speed. A 1-pole LPF (0.5–8 Hz cutoff, also driven by Mod Rate) bandlimits the output to the sub-audio range. Adaptive RMS normalization keeps the modulation signal approximately within [-1, 1] regardless of the chaotic state.

Hybrid Interpolation: The delay line uses two interpolation strategies selected by delay length. Below 10ms (or fewer than 2 samples), a 3rd-order Lagrange interpolator evaluates 4 sample points with Horner-form polynomial evaluation — preserves high-frequency content at short delays where precision matters most. At 10ms and above, a 1st-order Thiran all-pass filter (H(z) = (a + z⁻¹) / (1 + a·z⁻¹)) handles fractional delay with flat group delay and minimal CPU cost. The Thiran coefficient a is computed from the fractional delay part. Filter state is reset on transition from Lagrange to prevent clicks.

Phase Alignment (HQ Mode): In HQ mode (slider 5), a compensatory Thiran all-pass is applied to the Mid channel using the same fractional delay as the Side path. This ensures group delay matching between the two paths, maintaining phase coherence at the expense of additional latency. In Live mode, the Mid channel passes through unprocessed and the Haas effect provides the spatial perception — zero added latency but with slight phase offset between channels.

Width Control: The delayed Side signal is scaled by p_width × 2 and added to the original Side. At 0%, no delayed side is added (bypass). At 50%, the delayed side is at unity relative to the original. At 100%, the delayed side is 2× the original, creating maximum stereo expansion. The Base Delay sets the center delay time (5–35ms), and Mod Depth controls how far the modulation deviates from the base delay (as a percentage of the usable range between 5ms and 35ms).

Signal Flow: Input → M/S Encode → Mackey-Glass DDE modulation → Delay line with hybrid interpolation → Phase alignment (HQ: Thiran on Mid) → Width blend (original Side + modulated Side) → M/S Decode → Dry/Wet mix → Anti-aliasing LPF (~16kHz, exponential decay) → Output. A fade-in ramp (10ms) prevents clicks on initialization or parameter changes.

Lozengrad Pedal is a multi-pedal distortion plugin that models 7 distinct guitar pedal circuits, each with its own clipping topology, tone shaping network, and component-level EQ. It integrates a shared pre-processing stage and a global post-EQ (Low/Mid/High) that works across all modes.

Shared Signal Chain: All pedal modes pass through Input Level → Pre-EQ HPF (80Hz, first-order) → DC Blocker (0.997 alpha) → [pedal-specific processing] → Anti-Alias LPF (at 0.48× sample rate, Q=0.707) → Tone Stack (peaking + low-shelf, tone-dependent) → Post-EQ (Low peaking @ 100Hz, Mid peaking at sweepable frequency, High peaking @ 5kHz) → DC Blocker → 0 dBFS limiter. The post-EQ is shared across all modes with independent ±15dB per band and a sweepable mid frequency (200–5000 Hz).

Vintage Clip (mode 0): Models a classic two-stage op-amp clipper. Signal passes through soft clipping (g · (27 + g²) / (27 + 9·g²) with DC bias) followed by hard clipping (g / (1 + g⁸)^(1/8) at 3× pre-gain). The tone stack uses two filters: a peaking EQ (400–1000 Hz, −6 to 0 dB) and a low-shelf (2–4 kHz, 0 to +8 dB), both interpolated by the Tone control.

Turbo Clip (mode 1): Extends Vintage Clip with a mid-boost stage between the two clippers. After soft clip, the signal passes through a peaking filter at 900 Hz (Q=1.0, +6 dB), a notch at 2500 Hz (Q=2.0, −6 dB), and a low-shelf at 200 Hz (Q=0.7, −3 dB) — modeling the frequency-shaping network of a turbo distortion pedal. Final output goes through a second soft clip followed by hard clip.

Orange Drive (mode 2): Models a classic two-stage distortion. Soft clipping is followed by hard clipping at ±0.7 threshold. The tone network blends a 720 Hz LPF and 1060 Hz HPF (both first-order), where the Tone control crossfades between the two — full counter-clockwise = low-pass (dark), full clockwise = high-pass (bright). The blend is scaled by 1.4× post-gain.

Green Drive (mode 3): Models a classic overdrive with asymmetric soft clipping (g / √(1 + g²) with separate positive/negative gain). The clipped signal is blended with dry at a 70/30 ratio (the classic parallel mix). Tone control drives a peaking EQ at 1500 Hz (±10 dB). Output passes through a 47pF LPF (25 kHz, modeling the C4 capacitor) and a 5 kHz LPF (modeling op-amp bandwidth limiting).

Slam Boost (mode 4): Models a clean boost/overdrive with a tight front-end character. First-order HPF at 300 Hz removes low-end mud, followed by a peaking EQ at 1000 Hz (Q=1.0, ±10 dB via Tone). No clipping stage — pure EQ shaping with post-gain. Designed as a front-end boost for tightening high-gain amp tones.

Modern Drive (mode 5): Models a precision drive architecture with tight low-end and aggressive midrange. First-order HPF at 400 Hz tightens the low end, then soft clipping (g / √(1 + g²) with Gain pre-gain). A peaking EQ at 800 Hz (Q=1.5, +4 dB) provides mid presence, and a second peaking at 1000 Hz (Q=1.0, ±10 dB via Tone) acts as a "Bright" control. Post-gain of 4× compensates for HPF and clipping losses.

Grind Distortion (mode 6): Models a dual-gyrator distortion with aggressive mid-range shaping. First-order HPF at 35 Hz removes DC/subsonic content. Two gyrator-based peaking filters (simulated inductor EQ) at 250 Hz (Q=2.0, gain proportional to distortion control) and 1 kHz (Q=2.0) shape the mid-range before clipping. Soft clipping (g / √(1 + g²)) followed by hard clipping at ±0.4 threshold creates the signature aggressive distortion. Post-gain of 2×.

Oversampling (slider 5): Selectable 4× or 8× oversampling (not yet implemented in the processing path, slider is prepared for future anti-aliasing enhancement).

GFX Display: Two response curve plots. The top plot shows the pedal-specific tone stack frequency response (dependent on pedal mode and Tone slider). The bottom plot shows the shared 3-band Post-EQ response (Low/Mid/High). Both plots span 20 Hz–20 kHz with a center 0 dB reference line, and the curves update in real-time as parameters change.

Lozengrad RetroSpec is a purely cosmetic FFT spectrum visualizer with 11 absurd visual styles and 12 color themes. It does absolutely nothing to your audio — it passes through untouched while displaying your music as ASCII art, hex dumps, Chinese characters, or even a smiley face. Made for fun, not for mixing. ext_tail_size = -1 ensures REAPER knows this plugin adds zero latency and zero processing.

Under the hood: A 2048-point FFT with Hann windowing runs on the mono-summed input. Magnitude bins are smoothed (fast attack 0.5, slow decay 0.95) with peak hold (0.993 decay) and a 4.5 dB/oct slope compensation for a natural-looking spectrum. But honestly, none of that matters — you're here for the eye candy.

Visual Styles (slider1): Terminal — classic ASCII block bars (#, *, .). Blocks — Unicode block shades (█, ▓, ▒, ░). Matrix — random hexadecimal digits raining down like the Matrix. Retro — spinning animation (|, /, -, \) for that old-school modem feel. Mask: Smile — the spectrum reveals a smiling face hidden in the noise. Mask: Crescent — a crescent moon and star appear from the frequencies. Mask: Skull — your music draws a skull. Because why not. Radar — polar radar display with concentric circles and a + at the center. Hanzi — Chinese characters scaling from 一 (one) to 龘 (dragon flight) as the signal gets louder. Wormhole — concentric pulsing rings that expand and contract with the spectrum. HexDump — completely random hex digits pretending to be meaningful data.

Color Themes (slider2): 13 carefully curated themes: Gruvbox, Tokyo Night, Dracula, Matrix (green-on-black), Cyberpunk, Nord, Solarized, Monokai, One Dark, Catppuccin, Ayu, Everforest, and Rose Pine. Each theme defines 10 color slots (bg, bg0, fg, fg2, green, green2, yellow, red, aqua, orange) mapped to the spectrum's amplitude gradient.

Does it make your music sound better? No. Does it look cool? Absolutely.

Lozengrad Spektra is an FFT-based intelligent spectral density controller — a spectral compressor that builds a long-term EMA profile of the signal, detects excess energy via local contrast analysis with psychoacoustic weighting, and applies per-bin gain reduction with bidirectional frequency smoothing and per-bin attack/release. It is designed for transparent spectral balancing, resonance suppression, and tonal shaping.

FFT Engine: Uses a 2048-point FFT with 512-sample hop size (4× overlap) and Hann windowing, providing 1025 frequency bins with WOLA (Weighted Overlap-Add) reconstruction. The FFT processes linked-mid stereo (average of L and R magnitudes per bin) for spectral analysis but applies per-channel gain independently. Plugin delay compensation is reported as 2048 samples.

EMA Spectral Profile: A per-bin exponential moving average tracks the long-term spectral envelope. The Profile Down time constant (100–8000ms) controls how quickly the profile adapts when the signal level drops — slower settings create a more static "ceiling" that suppresses any excess above the learned spectrum. Profile up time is fixed at 120ms for fast tracking of level increases. The per-bin ratio of current magnitude to profile magnitude gives the excess profile, the primary detector signal.

Local Contrast Analysis: A bidirectional frequency-smoothed floor is computed for each bin (forward-backward one-pole filter with selectivity-dependent coefficient). The ratio of current magnitude to this local floor gives the contrast excess — detecting spectral peaks that stand out from their neighbors regardless of absolute level. The final detector blends profile excess and contrast excess via Selectivity (slider 5), where 0% = pure profile tracking, 100% = pure local contrast.

Psychoacoustic Weighting (slider 6): A perceptual weight curve based on Gaussian octave filters at 45Hz (low guard, −0.34), 280Hz (mud, +0.10), 3.3kHz (presence, +0.28), and 12.5kHz (air, −0.12). These model the ear's sensitivity to different frequency regions. Ear Weight controls the blend: at 0%, psychoacoustic weighting is flat; at 100%, the full perceptual model is applied, making the detector more sensitive to presence region excess and less sensitive to subsonic rumbles and extreme highs.

Adaptive Threshold (slider 14): The per-bin threshold is computed as threshold = (1.55 − 0.45 × sensitivity) × freq_bias, where freq_bias further adjusts based on psychoacoustic weight. Threshold (%) shifts the base threshold: 100% = nominal, 0% = much more sensitive (more reduction), 200% = much less sensitive (less reduction). Only bins exceeding their threshold trigger gain reduction, and the excess depth is normalized into a 0–12 range for consistent GR behavior across different signal levels.

Spectral Gain Computation: Per-bin gain reduction is computed as GR = −depth × (norm / (norm + knee))², giving a soft-knee response where the knee is fixed at 0.5. Depth (0–48dB) sets the maximum reduction per bin. A per-bin suppress weight from the interactive HP/LP/Bell filters additionally scales the gain reduction — bins outside the passband or in the bell boost/cut region get modified suppression.

Bidirectional Smoothing: Gain is smoothed in two passes. Frequency smoothing (forward-backward one-pole filter, coefficient determined by Smoothness 0–100%) ensures neighboring bins have coherent gain, preventing spectral holes. Temporal smoothing applies per-bin attack/release, where attack (~5–200ms) and release (~50–2000ms) are both proportional to Speed (slider 9). Fast speed = quick response, slow speed = gentle, stable reduction.

Transient Protection: A fast/slow envelope follower tracks the input peak. The transient amount (ratio of fast to slow envelope) modulates the effective depth, reducing gain reduction during transient events to preserve attack. This prevents spectral compression from dulling drum hits and percussive elements.

Interactive GFX Display: A real-time spectrum analyzer with per-bin before (cyan) and after (green) bars, plus red gain reduction bars from the top. Frequency axis spans 20Hz–Nyquist with octave markers and 12dB vertical grid lines. Three draggable filter handles control the per-bin suppress mask — HP (high-pass, 20Hz–20kHz), LP (low-pass, 20Hz–20kHz), and Bell (peak/cut, ±12dB, Q-controlled). Double-click any handle to reset. Filter state persists across bypass/transport via gmem. Input/output metering, GR peak, detector readings, and transient percentage are displayed in the footer.

Output Stage: After dry/wet mix, auto makeup (proportional to average GR), and output gain, a drive stage and soft-clipper provide final level control. The soft-clipper uses a cubic polynomial transition between 75% threshold and the ceiling (−12 to 0dB), with a clip indicator in the display. Clip Drive (0–100%) adds pre-clipper gain up to 1.5× for aggressive limiting.

Lozengrad Spyke is a transient shaper with independent Attack and Sustain controls, designed for quick and effective dynamic shaping. It uses three envelope followers to detect and separate transient peaks from the sustain tail, then applies gain modulation to either emphasize or suppress each component.

Envelope Detection: The detector works on the energy signal rect = (L² + R²)/2 processed through three parallel envelope followers. A fast envelope (1ms attack, 50ms release) tracks the instantaneous transient peaks. A slow envelope (15ms attack, 50ms release) tracks the body or average level. A sustain envelope (15ms attack, 200ms release) tracks the lingering tail with its long release time.

Transient Detection: The transient component is isolated as att_diff = max(fast_env − slow_env, 0) — when the fast envelope exceeds the slow envelope, a transient is present. This difference is normalized by the slow envelope amplitude (att_env = att_diff / slow_env) to ensure level-independent behavior. The Attack control (−100% to +100%) applies this normalized envelope as gain modulation: positive values boost transients at 2× aggression for emphasis, negative values reduce them for softening.

Sustain Detection: The sustain component is sus_diff = max(sustain_env − fast_env, 0) — when the long-release sustain envelope exceeds the fast transient envelope, the signal is in the decay/tail portion. Normalized as sus_env = sus_diff / fast_env. The Sustain control (−100% to +100%) modulates this: positive values boost at 1.5× for sustain enhancement, negative values reduce at 0.5× for tightening.

Gain Modulation: The total modulation factor is mod = 1.0 + att_env · att_gain + sus_env · sus_gain, clamped to a minimum of 0.05 to prevent polarity inversion. The output is input × mod × output_gain, followed by a safety clipper at ±0.9886 (−0.1 dBFS).

Lozengrad Trafo is a dual-path transformer saturation plugin with ADAA (Anti-Derivative Anti-Aliasing) on both paths. It splits the signal into two parallel nonlinear processors — an asymmetric Nickel path optimized for transient attack, and a symmetric Iron path with hysteresis feedback for sustain — then dynamically morphs between them based on transient content. The result is a saturator that responds to playing dynamics like a real transformer.

Nickel Path (Asymmetric): Models the asymmetric saturation of a nickel-core transformer, where positive and negative half-cycles distort differently. The transfer function antiderivative is F(x) = x²/2 + 0.05x³ − 0.0825x⁴, producing a polynomial with even-order harmonics. ADAA is applied via y = (F(xₙ) − F(xₙ₋₁)) / (xₙ − xₙ₋₁), with a derivative fallback when the difference is near zero. A fixed makeup gain of 1.5× ensures Nickel path audibility against the Iron path's hysteresis boost. Drive range: 0.5–8.0, clamped to ±1.5.

Iron Path (Symmetric + Hysteresis): Models the symmetric saturation of an iron-core transformer with magnetic hysteresis. The antiderivative is F(x) = x²/2 − 0.0555x⁴, a purely symmetric polynomial (odd-function derivative, only odd harmonics). Hysteresis is implemented as a low-passed feedback path: x_h = x_raw + 0.3 · hyst where hyst is a 250Hz one-pole filtered version of the input. This creates a frequency-dependent phase shift and saturation memory effect — higher frequencies experience more hysteresis smearing. Drive range: 0.5–10.0, clamped to ±1.5.

Transient-Driven Dynamic Morph: A fast envelope follower (instant attack, 10ms release) with sensitivity-gated noise floor tracks the input for transient detection. The derivative of this envelope is scaled by 5 + sensitivity × 15 to produce cv_rate [0,1]. A sharpening function maps this to cv_sharp (threshold at 0.05, saturation at 0.5), and the Morph Curve parameter (0.3–3.0) applies morph_cv = cv_sharp^(1/morph_c). Lower curve values make the morph more binary (hard switch), higher values make it more gradual. The Nickel weight is sqrt(morph_cv) and Iron weight is sqrt(1 − morph_cv), each with independent release envelopes (~42ms) for smooth crossfading. Weights are normalized to sum to unity.

Envelope Follower: A separate envelope follower with user-controlled Attack (0.1–20ms) and Release (10–500ms) tracks the input RMS to produce a control voltage. The CV is computed as CV = min(env / (1 − sensitivity × 0.999), 1.0) — higher sensitivity lowers the threshold, making the compressor-like gain reduction easier to trigger. This CV is smoothed (1ms time constant) and used in the scope display's activity boost but not in the morph itself (which uses the separate fast transient detector).

Auto Gain & Clip: Auto Gain tracks input and output RMS with 0.9995 envelope coefficients (~2000-sample time constant) and computes a makeup gain clamped to [0.25, 4.0] (±12dB range). The Clip option applies a hard 0 dBFS limiter after output gain.

GFX Scope: A real-time waveform oscilloscope (48,000-sample buffer, 4-sample stride = ~1 second at 48kHz) with per-pixel min/max accumulation for crisp rendering. The L channel is drawn at full opacity, R channel slightly dimmer and offset by 1 pixel. Each pixel column is colored based on the Nickel/Iron blend — cyan when Nickel dominates (transient attack), orange when Iron dominates (sustain). The color intensity (activity) scales with the envelope CV and per-path saturation level, so quieter signals appear white/grey and heavily saturated signals show full color. A HUD at the top displays all parameter values and diagnostic readings.

Lozengrad Hudut (Turkish: border / limit) is a true peak brickwall limiter with multi-rate polyphase oversampling, dynamic release, and a harmonic inflator stage. It detects and attenuates inter-sample peaks by operating on an oversampled signal, with up to 8× internal processing for maximum true peak accuracy.

Oversampling (slider4): Selectable 1×, 2×, 4×, or 8× using a multi-rate polyphase cascade. Each 2× stage consists of zero-stuffing (with ×2 gain compensation for the first sample) followed by a 23-tap half-band FIR filter exploiting the half-band property (all even-indexed coefficients except c0 are zero). The upsampling and downsampling chains are implemented as nested loops — each stage operates at its own internal rate, and downsampling uses the same HBF in reverse order, keeping only every other output sample. An HBF gain compensation factor (1.06 at 4×, 1.12 at 8×) corrects level differences across OS modes.

Signal Flow: Input → [Optional Harmonic Inflator] → Sidechain HPF → Envelope Detection → Gain Computation → Lookahead Delay → Gain Apply → Makeup → Hard Clip → Output. The entire gain reduction chain (envelope, gain computer, delay, application) runs inside the oversampled loop so that attenuation decisions are made at the true peak resolution.

Sidechain & Envelope: The sidechain passes through a first-order IIR HPF (20–250 Hz, slider9) to remove low-frequency content from the gain reduction detection, allowing the limiter to ignore sustained bass and react primarily to transient peaks. The envelope follower uses an attack time of 0.01–5 ms (slider8) computed as a first-order IIR: att_coeff = 1 − exp(−1 / (attack_s · os_srate + 1)).

Dynamic Release: The release time is modulated by the crest factor of the sidechain signal, computed from two parallel RMS followers — fast RMS (15 ms window) and slow RMS (120 ms window). The dynamic release scales the base release (10–500 ms, slider7) by 0.5× to 3.0× using min(max(1 / crest, 0.5), 3.0). This means percussive material (high crest) gets faster release, while dense material (low crest) gets slower, more transparent release. A hold timer equal to the lookahead duration holds the gain reduction at its minimum between peak detections.

Lookahead (slider3): 0–10 ms lookahead implemented as a circular delay buffer (up to 480,000 samples). Separate write and read pointers track the delay offset, with the delay length scaled by the OS factor. The buffer is reset on playback start to prevent stale-sample clicks.

Harmonic Inflator (slider5, slider6): A harmonic enhancer placed before the limiter detection. Two modes: Transparent uses a symmetric polynomial x · (1 + amount · (1 − x²)) for even-order harmonics with minimal tonal shift; Color applies the same polynomial with a DC offset (x + amount · 0.2) then subtracts the DC component, producing asymmetric even harmonics with a warm, vintage character.

Makeup & Ceiling: After gain reduction, the signal is multiplied by ceiling / threshold makeup gain, then hard-clipped to ±ceiling_amp. The ceiling ranges from −14 to 0 dB (slider2) and the threshold from −24 to 0 dB (slider1). Plugin Delay Compensation is reported as lookahead + FIR group delay (0/11/16/19 samples for 1×/2×/4×/8×).

GFX Display: A 540×380 scrolling waveform area shows peak amplitude as a green waveform (dB scale, −60 to 0 dB with grid lines at 0, −6, −12, −18, −24, −36). Gain reduction is overlaid as a red bar graph falling from the top, with a smoothed GR readout in the bottom-right corner. Both audio and GR traces use a dual-layer rendering (wide low-opacity glow + narrow high-opacity core) for visual clarity. Data is decimated to ~250 Hz update rate and stored in 600-bin ring buffers.

Lozengrad Guitarbus One is an adaptive guitar bus processor with a single Intensity knob. Everything else — compression, saturation, EQ, multiband compression, harshness control, limiting — is continuously derived from real-time signal analysis of chug density, palm mute ratio, pick attack harshness, sustain/staccato ratio, and body energy. The result is a plugin that mirrors the behavior of a full guitar bus processing chain, adapting to every riff and chord change.

Signal Flow: Input → Analysis → Decision Engine → HPF → Adaptive Mud Suppression (200–400Hz dynamic cut) → Feedback Bus Compressor → Mid-Only Tube Saturation (300–800Hz) → Precision EQ (7-band, evolving from transparent to shaped) → 4-Band Multiband Compressor (LR4 crossover: chug/body/presence/pick) → Dynamic Harshness Control (4–7kHz de-esser) → Safety Limiter → Auto Gain → Stereo Widen (bass mono) → Output.

One-Knob Architecture: The single Intensity slider (0–100%) nonlinearly activates each processing stage with staggered onset curves — Mud Suppression activates first (~15%), followed by the bus compressor (~30%), mid saturation (~25%), EQ (linear, 0–100%), multiband compressor (~40%), harshness control (~50%), and limiter (~80%). Each module's activation is further modulated by real-time analysis modifiers (crest factor, band energy ratios, transient density) and a confidence metric derived from crest factor stability. This creates an adaptive response that behaves differently on palm-muted chugs vs open chords vs lead lines, all from a single knob.

Mud Suppression: A dynamic 200–400Hz cut driven by an envelope follower on the band energy. As chug density increases, the cut deepens (up to 4 dB), keeping the low-mid range clean and tight even during heavy palm-muted passages.

Bus Compressor: A feedback-style VCA compressor with RMS detection, soft knee, and program-dependent release. Threshold and ratio are both derived from the adaptive engine — at higher intensity settings, threshold drops and ratio increases for more aggressive glue compression. Side and Mid channels get independent threshold offsets.

Mid Saturation: A tube-style saturator applied only to the 300–800Hz band, targeting the core body of the guitar tone. Drive increases with intensity, adding harmonic richness to the midrange where guitars live, while leaving the lows and highs clean.

Precision EQ: A 7-band biquad EQ that morphs from a transparent pass (10Hz HPF, flat mids, 20kHz LPF) to a guitar-specific shaping curve (60Hz HPF, scooped mids at 160Hz and 2.3kHz, presence at 2.8kHz, air cut at 3.5kHz, 9kHz LPF) as intensity increases.

4-Band Multiband Compressor: LR4 crossover splits the signal into Chug (<200Hz), Body (200–800Hz), Presence (800–2500Hz), and Pick (>2500Hz) bands, each with independent compression parameters derived from the adaptive engine. Thresholds drop and ratios increase with intensity, providing per-band dynamic control that targets the specific spectral regions of guitar playing.

Dynamic Harshness Control: A de-esser style processor targeting the 4–7kHz range. An envelope follower on the harshness band drives a dynamic cut of up to 6 dB at full intensity, taming excessive pick noise and fret buzz without dulling the overall tone.

GFX Display: A central intensity knob with a 34-dot LED ring (showing current value), input/output level metering, gain reduction readout, auto gain trim visualization via knob halo color (warm/cool), and a clip indicator. The knob supports click-drag, double-click-to-reset (50%), and mouse wheel. Two smaller knobs for input and output gain flank the main knob, with an auto gain toggle switch between them.

Lozengrad Drumbus One is an adaptive drum bus processor with a single Intensity knob. Like its guitar counterpart, it continuously analyzes the input signal — low-end energy, low-mid buildup, presence harshness, air content, crest factor, transient density, and stereo width — and derives per-module processing parameters in real time. The result is a drum bus that adapts to every kick hit, snare crack, and cymbal wash.

Signal Flow: Input → Analysis → Decision Engine → Adaptive Cleanup EQ → Adaptive Transient Shaper → Adaptive Compressor → Adaptive Saturation (tube/tape/console blend) → DC Blockers → Adaptive Tone EQ → Adaptive Clipper → Adaptive Limiter → Auto Gain → Stereo Widen (bass mono) → Output. All processing is done in Mid/Side.

One-Knob Architecture: The single Intensity slider (0–100%) activates each processing stage with staggered onset — Cleanup EQ activates earliest (~10%), followed by Transient Shaper (~30%), Compressor (~50%), Saturation (~45%), Clipper (~70%), and Limiter (~90%). Each module's activation is modulated by analysis-derived modifiers — crest factor, RMS level, band energy ratios, transient density, and stereo width — creating a response that adjusts to the material in real time.

Adaptive Cleanup EQ: Subtle corrective EQ using one-pole band-splitting filters. Four bands (low <120Hz, low-mid 150–350Hz, presence 3–6kHz, air 10–15kHz) receive gentle gain corrections based on how excessive each band is relative to the overall spectrum. Side channel gets half the correction to preserve stereo width.

Adaptive Transient Shaper: A fast/slow envelope follower pair detects transients. Attack mode boosts transient peaks (up to 35%), sustain mode reduces the body (down to –20%) — both scaled by the adaptive weight. This provides dynamic transient enhancement that automatically adjusts to the drum program.

Adaptive Compressor: A feed-forward glue compressor with RMS detection. Ratio ranges from 1:1 (bypassed) to 4:1, attack from 20ms to 8ms, release from 120ms to 60ms, and threshold from –12dB to –26dB — all proportional to the adaptive weight. The compressor tightens the drum bus naturally without manual threshold hunting.

Adaptive Saturation: A three-mode saturator (tube/tape/console) with frequency bias. When the low end is dominant, saturation is biased toward high frequencies to add presence without muddying the lows. Drive and high bias are both derived from the analysis.

Adaptive Tone EQ: Corrects the tonal shift introduced by saturation. A presence dip (2–5kHz) and air shelf (9kHz+) are applied based on the current saturation drive, maintaining clarity and air even at high saturation levels.

Adaptive Clipper & Limiter: A soft clipper with an adaptive threshold that follows the current peak envelope, and a transparent limiter that provides final ceiling control. Both scale with the adaptive weight, providing peak protection that tightens automatically as the drum bus gets hotter.

Smart Auto Gain: Toggleable RMS-based loudness matching (200ms window) that compares input and output levels and applies up to ±6dB of trim gain, ensuring consistent perceived loudness regardless of how much processing is applied.

GFX Display: Identical layout to Guitarbus One — a central intensity knob with 34-dot LED ring, input/output metering, gain reduction readout, auto gain trim visualization via knob halo color, and a clip indicator. Two smaller knobs for input and output gain flank the main knob, with an auto gain toggle switch between them.

Lozengrad Masterbus One is an adaptive mastering bus processor with a single Intensity knob. It analyzes the full mix — crest factor, RMS level, dynamic range, spectral balance, and transient density — and derives per-module processing parameters in real time. Designed for transparent mix bus processing that adapts to every section of the song.

Signal Flow: Input → Analysis → Decision Engine → Tape Saturation → Passive Tube EQ → Soft Clipper → Bus Compressor → Precision Mastering EQ (static shaped + dynamic bands) → True Peak Limiter → Auto Gain → Output. All processing is pure stereo L/R (no mid/side matrix).

One-Knob Architecture: The single Intensity slider (0–100%) activates each stage with staggered onset — Tape Saturation activates earliest (~15%), followed by Tube EQ (~25%), Soft Clipper (~40%), Bus Compressor (~50%), Mastering EQ (~60%), and Limiter (~80%). Each module's activation is modulated by crest factor, RMS level, spectral balance, and confidence metrics, creating a response that behaves differently on quiet verses vs loud choruses.

Tape Saturation: A gentle tape-style saturator with pre-emphasis (1.5kHz high-shelf) before saturation and de-emphasis after. Uses the tanh approximation with drive normalized by tanh(d) / d for consistent output level. At low intensity, the saturation is nearly transparent; at higher settings, it adds subtle harmonic warmth and density to the entire mix.

Passive Tube EQ: A two-band EQ modeled after passive tube equalizer circuits. A low shelf at 120Hz (Q=0.6) and a high shelf at 8kHz (Q=0.6), both with gain proportional to the adaptive weight. The EQ uses RBJ biquad coefficients recomputed on every parameter change, with state attenuation to prevent zipper noise. Designed for broad musical shaping rather than surgical correction.

Soft Clipper: A gentle soft-clipping stage using a polynomial transfer curve. Provides subtle peak rounding before the compressor, reducing the workload on downstream stages. The clip threshold adapts to the current signal level, engaging more aggressively on transient peaks while leaving lower-level material untouched.

Bus Compressor: A VCA-style bus compressor with RMS detection, soft knee, and program-dependent release. Threshold and ratio are derived from the adaptive engine, providing gentle glue compression that tightens the mix without audible pumping. The compressor includes a 60Hz sidechain high-pass filter to prevent bass from over-triggering gain reduction.

Precision Mastering EQ: A dual-mode equalizer with static shaped bands and dynamic bands. Static bands provide a gentle mastering curve with broad shelves and presence shaping. Dynamic bands respond to the adaptive engine, providing spectral correction that adapts to the material — adding air when the mix is dull, taming harshness when the highs are aggressive.

True Peak Limiter: A lookahead limiter with oversampled true peak detection. Provides transparent ceiling control with program-dependent release that responds to crest factor. The limiter is the final safety stage, catching intersample peaks while maintaining the dynamics shaped by the preceding stages.

GFX Display: A central intensity knob with 34-dot LED ring, input/output metering, gain reduction readout, auto gain trim visualization, and a clip indicator. Two smaller knobs for input and output gain flank the main knob, with an auto gain toggle switch between them.

Lozengrad Synthbus One is an adaptive synth bus processor with a single Intensity knob. It analyzes sub energy (30–100Hz), resonance/screech energy (1.5–4.2kHz), air/digital-edge energy (6–10kHz), stereo width, crest factor, and attack density (plucks/arps vs sustained pads). The result is a bus processor that adapts to every patch — from warm pads to screeching leads to massive supersaws.

Signal Flow: Input → Analysis → Decision Engine → Sub Anchor (mono sum + low-end weight) → Resonance Tamer (filter-screech control) → Glue Compressor → Ensemble Widener (unison/chorus thickening) → Harmonic Warmth → Precision EQ → 4-Band Multiband Compressor → Limiter → Auto Gain → Output.

One-Knob Architecture: The single Intensity slider (0–100%) activates each stage with staggered onset — Sub Anchor activates first (~10%), followed by Resonance Tamer (~20%), Glue Compressor (~35%), Ensemble Widener (~45%), Harmonic Warmth (~50%), Precision EQ (~55%), Multiband Compressor (~65%), and Limiter (~80%). Each module's activation is modulated by synth-specific analysis metrics.

Sub Anchor: Mono-sums the low end below a crossover point (derived from sub energy analysis) and applies a gentle low-shelf boost to anchor the synth's fundamental. Prevents low-end phase cancellation while maintaining a solid sub foundation.

Resonance Tamer: Targets the 1.5–4.2kHz range where filter resonance, sync sweep harshness, and digital edge artifacts live. A dynamic cut driven by an envelope follower on the resonance band reduces harshness by up to 5dB without dulling the overall tone.

Glue Compressor: A feed-forward compressor with RMS detection and program-dependent release. Threshold, ratio, attack, and release are all derived from the adaptive engine, providing compression that adapts to the synth's dynamic profile — gentle on pads, tighter on plucks and arpeggios.

Ensemble Widener: A stereo width enhancer that simulates unison/chorus thickening. Analysis of the side-to-mid ratio drives adaptive widening — narrow patches get more width, already-wide supersaws get subtle tightening to maintain mono compatibility. A 180Hz low-pass on the side channel keeps bass centered.

Harmonic Warmth: A tube-style saturator biased toward the low-mids (200–800Hz). Adds even-order harmonics to thin-sounding patches while leaving dense, rich patches clean. The saturation blend is modulated by the adaptive engine based on spectral balance analysis.

Precision EQ & Multiband Compressor: A 7-band precision EQ and 4-band LR4 multiband compressor adapted from the guitar bus engine, with synth-specific frequency splits and parameter ranges. The EQ provides broad spectral shaping, while the multiband comp targets per-band dynamics for sub, body, presence, and air bands.

GFX Display: A central intensity knob with 34-dot LED ring, input/output metering, gain reduction readout, auto gain trim visualization, and a clip indicator. Two smaller knobs for input and output gain flank the main knob, with an auto gain toggle switch between them.

Lozengrad Bassbus One is an adaptive bass bus processor with a dual-zone single Intensity knob. It splits the intensity range into two zones: 0–50 engages the full adaptive clean processing chain, while 50–100 keeps the clean chain maxed out and introduces a Grit + Slam parallel compression layer for aggressive, coloured bus-slam character. Analysis targets low-end energy, mid-range punch, pick/string noise, sustain, crest factor, and transient density.

Signal Flow: Input → Analysis → Decision Engine → Low-Mid Mud Suppression → Feedback Bass Compressor → Sub Weight Enhancer → Harmonic Saturation → Precision EQ → 4-Band Multiband Compressor → String Noise / Fret Buzz Control → Grit + Slam Parallel Compressor → Limiter → Auto Gain → Output.

Dual-Zone Architecture: The Intensity knob (0–100%) is split into two operating zones. Zone 1 (0–50): the full adaptive clean chain ramps up, reaching full activation by the halfway point — providing transparent bus processing that adapts to the bass performance. Zone 2 (50–100): the clean chain stays fully engaged while a parallel Grit + Slam compression layer ramps in. This layer uses a crushed parallel compressor blend with heavy saturation, providing the aggressive bus-slam character without losing the clean foundation. The transition is seamless and continuous.

Low-Mid Mud Suppression: A dynamic 200–400Hz cut driven by an envelope follower on the mud band. As low-mid buildup increases (common with palm-muted bass or aggressive picking), the cut deepens to maintain clarity and definition in the upper bass register.

Feedback Bass Compressor: A feedback-style compressor with RMS detection optimized for bass. Uses a 60Hz sidechain high-pass filter to prevent sub frequencies from over-triggering. The compressor provides smooth, musical gain reduction that adapts to the bass player's dynamics — tighter on busy passages, looser on sustained notes.

Sub Weight Enhancer: A sub-frequency processor that analyzes the 40–100Hz range and applies dynamic low-shelf boost or harmonic enhancement. When the sub energy is low, the enhancer adds weight through gentle saturation; when the sub is already prominent, it provides subtle tightening to prevent muddiness.

Harmonic Saturation: A tube/tape blend saturator targeting the upper bass and low-mid range (100–800Hz). Adds presence and grind to the bass tone, helping it cut through the mix without excessive high-frequency content. The saturation blend adapts to the playing style — cleaner on fingerstyle, more aggressive on picked or slapped bass.

String Noise / Fret Buzz Control: A de-esser style processor targeting the 2–5kHz range where string noise, fret buzz, and finger squeaks live. A dynamic cut of up to 4dB tames unwanted mechanical noise while preserving the attack and presence of the bass.

Grit + Slam Parallel Compressor: Active in Zone 2 (50–100% intensity). A parallel compression bus with heavy ratio (8:1+), fast attack, and aggressive saturation. The crushed signal is blended with the clean chain, with the blend amount proportional to the intensity above 50%. At full intensity, the parallel bus adds significant density and aggression while maintaining the fundamental low-end clarity from the clean chain.

GFX Display: A central intensity knob with 34-dot LED ring, input/output metering, gain reduction readout, auto gain trim visualization, and a clip indicator. Two smaller knobs for input and output gain flank the main knob, with an auto gain toggle switch between them.