Punch

Limiter — Position 6

Punch is a two-stage loudness processor:

1. Transient Shaper — restores or enhances attack characteristics that preceding compression may have dulled (applied pre-clip)
2. Clipper — eliminates intersample peaks and creates headroom for loudness maximization

The key design decision: transient shaping runs pre-clip. Post-clip shaping would create time-varying gain modulation that sounds like pumping on every note attack. Pre-clip shaping lets the clipper naturally limit the enhanced peaks.

Quick Start

For loudness maximization with preserved punch: Clip Mode Soft, Threshold −1.0 dB, Softness 0.4, Oversampling 8x. Attack shaper at +0.2, Sustain at 0.0. This clips peaks cleanly while the transient shaper keeps attack feel intact.

Default Oversampling

The default is 4× oversampling. This balances CPU usage with aliasing rejection for most use cases. Use 8× for critical mastering work where maximum transparency is required.

---

Controls

Control	Range	Description
Clip Mode	Hard / Soft / Cubic	Waveshaper algorithm (see below).
Ceiling	−12 – 0 dBFS	Peak ceiling. −0.3 to −0.1 dBTP typical for streaming. −1.0 dB for mp3 intersample headroom.
Oversampling	1×, 2×, 4×, 8×	Upsample before clipping, downsample after. Higher = fewer aliasing artifacts, more CPU.
Transient Attack	0.0 – 1.0	Transient detection and upward expansion amount on attacks. Restores punch after compression.
Transient Sustain	0.0 – 1.0	Sustain portion enhancement. Lengthens decay of transient events.
Transient Release	0.0 – 1.0	Speed at which transient detection decays between events.
Mix	0.0 – 1.0	Dry/wet blend for the entire Punch stage. Allows parallel clipping (more transparent).
Bypass	On/Off	Bypasses all processing.

---

Clip Modes

Soft (tanh)

y = tanh(x / ceiling) × ceiling

Smoothest onset, most transparent. Begins softly limiting before the ceiling, adding gentle saturation as levels approach it. Recommended for vocal and acoustic buses, master bus processing, and anywhere transparency is paramount.

Cubic

Polynomial waveshaper. Falls between Hard and Soft:

• More headroom before onset than Hard
• Harder onset than Soft once ceiling is reached
• Adds odd-order harmonics (3rd, 5th) characteristic of solid-state saturation

Recommended default for most buses.

Hard

y = clip(x, -ceiling, ceiling)

Abrupt ceiling — maximum loudness, most aggressive character. Adds strong odd harmonics. Use for rock/metal buses where harmonic grit is appropriate.

---

Oversampling

Setting	Aliasing	CPU	Use Case
1×	High	Minimal	Fast iteration, monitoring only
2×	Moderate	Low	Rough mixes, stem passes
4×	Low	Moderate	Default — recommended for most work
8×	Very low	Higher	Critical mastering, maximum transparency

---

Techniques

Transparent Loudness Maximization

Control	Value
Clip Mode	Soft or Cubic
Ceiling	−0.3 dBFS
Oversampling	4×
Transient Attack	0.30 – 0.50
Mix	1.0

Set the ceiling so the meter just catches the loudest peaks. Then dial in Transient Attack to recover kick and snare punch lost to preceding compression stages.

Parallel Clipping

Mix at 0.70 – 0.85 to blend the clipped signal under the dry signal:

• Dry path preserves original transients
• Clipped path adds density and loudness
• More transparent than full wet at equivalent loudness levels

Streaming Targets

Platform	Target LUFS	Ceiling
Spotify	−14 LUFS	−1.0 dBTP
Apple Music	−16 LUFS	−1.0 dBTP
YouTube	−14 LUFS	−1.0 dBTP
Tidal	−14 LUFS	−1.0 dBTP
CD / Physical	Max loudness	−0.3 dBTP

Intersample Peaks

Set ceiling to −1.0 dBFS if the final delivery format is mp3 or AAC. Lossy encoding can create intersample peaks above the true peak of the source, and −1.0 dBFS provides sufficient headroom to avoid clipping in the encoder.

---

Implementation Notes

• Upsampling: linear interpolation with filter_state[1] tracking previous input across buffer boundaries
• Downsampling: simple average + very light IIR (pole = 0.05) — the low pole prevents the downsampler from contributing to pumping
• Transient detector uses the native sample rate, not the oversampled rate, to avoid detection artifacts
• Maximum oversampling factor: 16× | Maximum block size: 8192 samples

See Also

• Techniques & Presets — Punch clipping modes and loudness techniques
• Genre Signal Chains — how to use Punch per genre and target loudness
• Architecture — oversampling design and pumping-free transient shaper