VINAIclip — Manual

01Overview

VINAIclip is a final-stage clipper built for the very end of the chain — after your dynamics, before (or instead of) the limiter. It clamps the signal to a hard ceiling and turns the overshoot into harmonics rather than distortion you didn't ask for. The whole thing runs on the Transparent Ceiling Engine: a ceiling you can trust, with a morphing knee that goes from a clean brick-wall corner to a rounded, musical curve.

You pick the harmonic signature from three algorithms — Hard, Warm and Tube — set the threshold by ear with Delta listen soloing only what's being removed, and keep inter-sample peaks in check with a 4× true-peak safety clip. Oversampling runs up to 32× to keep aliasing out of the harmonics you just generated.

02The interface

• Scope (centre) — pre / post / delta clip view showing the flat-topped waveform and exactly how much is being shaved.
• Algorithm selector — Hard · Warm · Tube, each its own clipping kernel.
• Gain & clip controls — IN, OUT, THRESH and SOFT, with the LINK toggle that mirrors IN/OUT for instant makeup.
• Safety & quality — TRUE PK and the OVERSAMPLE selector (1×–32×).
• Metering (right) — LUFS / RMS loudness meter with gain-reduction indicator, plus the DELTA listen toggle.

03Knobs & buttons

Gain & clip

Safety, quality & metering

04The three algorithms

05Under the hood

At its core VINAIclip is a memoryless, per-sample waveshaper — it has no time constants, so it can clamp a transient peak instantly without the attack/release smear of a limiter. The "Transparent Ceiling Engine" is everything wrapped around that shaper: oversampling, anti-aliasing, true-peak detection and metering, all phase-linear on the audio path.

Signal flow

IN gain → upsample (polyphase FIR) → per-sample waveshaper (one of three curves, with the Softness knee blended in) → downsample/decimate (matching FIR) → optional 4×-oversampled true-peak clip → ~5 Hz DC blocker → OUT trim → meters & scope tap. With LINK on, the OUT trim tracks IN automatically so loudness-matched comparison is the default, not an afterthought.

• Waveshaper kernels — three static transfer curves applied per sample: a hard clamp (brick-wall limiter on amplitude), a symmetric tanh soft clip, and an asymmetric tanh that breaks even and odd harmonics differently for tube-style colour.
• Softness knee — a continuous 0–100% control that morphs the corner of the active curve from a razor-sharp brick wall to a progressively rounded knee, trading peak-catch precision for harmonic gentleness. Applied to every algorithm.
• Oversampling — polyphase-FIR up/downsampling at 1× / 2× / 4× / 8× / 16× / 32× pushes the harmonics generated by clipping above the audio band before they can fold back as aliasing. 32× is reserved for export (offline render); live playback caps at 16× to keep CPU sane. The FIR pair is linear-phase, so the clip stays phase-transparent regardless of oversampling factor.
• True-peak stage — an optional second clip running at 4× oversampling reconstructs the inter-sample waveform and clamps overshoots that sit between samples — the peaks a plain 0 dBFS sample meter never sees. It adds latency, which is reported to the host as plug-in delay compensation (PDC) so everything stays time-aligned.
• DC blocker — a one-pole high-pass at roughly 5 Hz on the output removes any DC offset the asymmetric Tube curve introduces, without touching the audible low end.
• Metering internals — loudness is measured as LUFS and RMS off the post-clip signal; the scope taps pre, post and delta in parallel so the delta path shows precisely the residual the shaper removed.
• Bypass — bypass is time-aligned against the active latency (oversampling + true-peak) so A/B is click-free and sample-accurate, never a jump in level or phase.

06Measured

Measured in-house (headless host + numpy/scipy), 48 kHz.

Head-to-head: VINAIclip vs StandardCLIP

Both clippers measured on the same bench, same host, same signal, identical settings — hard clip, softness 0, +6 dB drive into a 0 dB ceiling, true-peak off — with oversampling matched at every step. Peak alias component in dBc (relative to the fundamental); lower is cleaner.

Read the 9 kHz row the way a mix reads it: that is the cymbal / air band hitting the ceiling. At the shipping default, VINAIclip is 23 dB cleaner there than StandardCLIP at the same oversampling — and its 8× figure on the 2.5 kHz test (−108.8 dBc) is already below StandardCLIP's best at 32× (−106.4 dBc). Above 16× VINAIclip sits at the measurement floor of the bench itself, which is why 16× and 32× read within a couple of dB.

Test signals are FFT-bin-coherent sines (N = 65536 @ 48 kHz) so fundamental leakage cannot be misread as aliasing. StandardCLIP © SIR Audio Tools — independent product, no affiliation; measured from the public release build with its own oversampling set per step via saved plugin state.

07Specifications