Vocal Resonance Sculptor (S42)

AI-generated voices - from TTS models or LTX 2.3's built-in audio - tend to share one flaw: they're thin. The mids and highs are there, but there's no body, no chest. S42_VocalResonanceSculptor is the pack's attempt to fix exactly that: a harmonic exciter that adds low-register weight to a voice while keeping the top end smooth. If you've got a voiceover or generated vocal that sounds like it was recorded through a tin can, this is the node you reach for.

It's in the Audio/Enhancement category, and it does one thing well. The marketing says "chest" and "velvet," and both are accurate enough: it saturates the lows (chest) and tames the highs (velvet).

How it works

The chain is simple and legible: lowpass the vocal at chest_freq_hz × 1.5 to isolate the low register, run that through a tanh saturation stage scaled by chest_drive - tanh is a smooth, musical saturation that generates harmonics as you push it - then subtract the lows from the original to get the highs, soften those by velvet_smoothing, and sum both halves back together. The output is peak-normalized, so the louder, fuller voice doesn't clip.

In plain terms: it adds harmonic richness to the low end (the chest), then shaves a bit of harshness off the high end (the velvet), and recombines.

Inputs

• audio - the vocal waveform.
• chest_freq_hz - 110 default, 60–250. The crossover point; lower for deep voices, higher to bring out more of the mid body.
• chest_drive - 1.5 default, 0–5. How hard the low end gets saturated. 1–2 is a warm boost; past 3 it starts to distort noticeably.
• velvet_smoothing - 0.4 default, 0–1. Higher = softer, less sibilant top end; 0 leaves the highs as-is.

Output

A single AUDIO. No info, no meters - A/B it against the original and trust your ears.

Installing it

Part of S42 CutFlow: ComfyUI Manager → search "S42 CutFlow", or git clone https://github.com/GeekyGhost/S42-CutFlow into ComfyUI/custom_nodes/, install requirements, restart. Pure torch filter math - no models, no downloads, and it's fast.

Gotchas

Three honest notes. First, the hard peak-normalization at the end means the output's loudness can jump compared to the input - if the vocal suddenly sounds louder and fuller, that's by design, but plan your levels after it, not before. Second, chest_drive past ~2.5 on an already-hot signal gets muddy - the tanh saturates hard and you lose clarity, which is the opposite of the effect you wanted. Third, it's built for vocals; running a full music mix through it will pump the low end in ways you probably don't want. It also pairs naturally with the pack's DynamicRangeCompressor afterward if you want the enhanced vocal to sit glued into a mix.