| title | Satch |
|---|---|
| subtitle | User Manual |
| author | Michael Dungan |
| geometry | margin=2.5cm |
| colorlinks | true |
{ width=50% }
satch is a detail-preserving spectral saturator. It uses FFT-based spectral analysis to preserve quiet frequency components (detail) through the clipping process, producing textured flat-top clipping instead of the smooth, featureless flat tops of standard time-domain saturation.
Inspired by Newfangled Audio's Saturate.
Build from source (requires nightly Rust):
cargo nih-plug bundle satch --releaseThe bundler outputs to target/bundled/. Copy either the .vst3 or .clap file to your plugin directory:
- Linux:
~/.vst3/or~/.clap/ - macOS:
~/Library/Audio/Plug-Ins/VST3/or~/Library/Audio/Plug-Ins/CLAP/ - Windows:
C:\Program Files\Common Files\VST3\orC:\Program Files\Common Files\CLAP\
- Insert satch on a track
- Raise Gain to push the signal into clipping
- Turn up Detail to preserve quiet harmonics through the clipped sections
- Adjust Knee to taste (0% = hard clip, 100% = soft tanh)
Input gain boost. Range: 0 to +24 dB. Default: 0 dB.
Pushes the signal toward the clip ceiling. Higher gain = more of the signal exceeds the threshold and gets clipped. At 0 dB, the signal passes through at its original level (clipping only occurs if it already exceeds the threshold).
Clip ceiling. Range: -24 to 0 dB. Default: 0 dB.
Sets where clipping begins. At 0 dB, the ceiling is at full scale (±1.0). At -6 dB, the ceiling drops to ±0.5 -- any signal above that level is clipped. Below the threshold, the signal passes through unchanged.
Threshold is independent of Gain. Gain controls how hard the signal hits the ceiling; Threshold controls where the ceiling is.
Spectral detail preservation. Range: 0 to 100%. Default: 0%.
Controls how much quiet spectral content survives through the clipping process.
- 0%: Standard time-domain saturation. Flat tops are smooth and featureless -- all frequency content in the clipped region is destroyed.
- 100%: Full spectral detail preservation. Quiet harmonics ride through the clipped sections as visible texture on the flat tops. The peaks stay at the clip level; the texture hangs downward.
Detail only affects clipped portions of the signal. Unclipped material is identical at 0% and 100%.
The spectral analysis uses a 2048-point FFT with 75% overlap (512-sample hop), introducing 2048 samples of latency (compensated automatically by the host).
Clipping shape. Range: 0 to 100%. Default: 100%.
Controls the transition from linear to clipped:
- 0% (hard): Brick-wall clipping. The signal passes through linearly below the threshold and is hard-clamped at the threshold. Sharp transition, maximum edge.
- 100% (soft): Tanh saturation curve. Gradual transition into clipping -- signals approaching the threshold are gently compressed before reaching the ceiling. Warmer, rounder character.
- In between: Linear crossfade between hard and soft.
Dry/wet blend. Range: 0 to 100%. Default: 100%.
At 100%, the output is fully processed. At 0%, the output is the original dry signal (delayed to match latency). Use intermediate values for parallel saturation.
satch combines two clipping paths:
-
Time-domain path: Applies
threshold * tanh(gain * x / threshold)(crossfaded with a hard clip via the knee) to produce flat-top clipping at the threshold level. -
Spectral path: Uses an STFT (Short-Time Fourier Transform) to split each frame's bins into a loud path (bins within 6 dB of the spectral peak) and a quiet path (bins more than 20 dB below it, smoothly crossfaded between). The two paths are reconstructed by separate inverse STFTs; the nonlinear clipping is applied in the time domain after reconstruction (not per-bin in the frequency domain), which keeps the overlap-add normalization exact. The result has flat-top clipping character but with the quiet detail preserved.
The Detail knob blends between the two paths, but only in clipped regions. A clip-aware mask (tanh^2(gain * x / threshold)) detects where clipping is occurring and applies the spectral difference only there. Unclipped material always uses the time-domain path.
The output is safety-clamped at ±threshold × 1.5 -- detail texture can ride up to 50% above the threshold clip level rather than being limited exactly at it.
The window is freely resizable -- drag the plugin window's edge or corner in your host. (There are no in-plugin zoom buttons or keyboard shortcuts.) The scale tracks the window width and clamps to 0.5x-4.0x (50% - 400%); the chosen size is persisted across host restarts.
- Drag vertically on any dial to adjust (up = increase)
- Shift+drag for fine control (10x slower)
- Double-click any dial to reset to default
- Right-click any dial to type a value; Enter commits, Escape cancels, clicking outside auto-commits
- No audio-thread allocations -- the process() callback never allocates heap memory in release builds
- CPU rendering -- uses tiny-skia (software rasterizer) + fontdue (glyph cache) + softbuffer (pixel buffer). No OpenGL context, no GPU drivers loaded
- Time-domain nonlinearity -- the tanh clipping is applied to the reconstructed signal in the time domain (after the two inverse STFTs), not per-bin in the frequency domain; this keeps the overlap-add COLA normalization exact. Quiet bins (detail) pass through while the loud fundamental is clipped
- Clip-aware detail blend --
tanh^2mask ensures spectral detail is only applied where the signal is actually being clipped, leaving unclipped material identical to the time-domain path - Peak-relative spectral split -- bins within 6 dB of the spectral peak are classified as "loud"; bins more than 20 dB below are "quiet"; smooth crossfade in between
Benchmarks (Bitwig, 48 kHz / 1024 samples, GUI closed):
| Instances | CPU | RSS | Per Instance |
|---|---|---|---|
| 100 | 13.7% | 82 MB | ~0.82 MB, 0.14% CPU |
- CLAP
- VST3
- Standalone (JACK, or CPAL -- ALSA on Linux, CoreAudio on macOS, WASAPI on Windows -- with a dummy/offline fallback)
GPL-3.0-or-later