ComfyUI_BiSpectrum

ComfyUI custom node that computes the Bispectrum (triple correlation spectrum) of an audio signal and outputs it as an image.

---

What is the Bispectrum?

The bispectrum is a higher-order spectral analysis tool defined as:

$$B(f_1, f_2) = \mathbb{E}\bigl[X(f_1)\cdot X(f_2)\cdot X^*(f_1+f_2)\bigr]$$

where $X(f)$ is the discrete Fourier transform of the signal. It reveals nonlinear frequency interactions (phase coupling) between pairs of frequency components — information that is invisible in the ordinary power spectrum.

---

Node

Audio to Bispectrum

Input	AUDIO (ComfyUI native audio type)
Output	IMAGE — 2-D bispectrum plot [1, H, W, 3]
Category	audio

Parameters

Parameter	Default	Range	Description
nfft	512	64 – 1024	FFT size. Larger → finer frequency resolution, slower
hop_length	256	32 – 1024	Frame shift in samples
win_length	512	64 – 1024	Analysis window length in samples
normalize	OFF	ON / OFF	ON: bicoherence² ∈ [0, 1] / OFF: raw magnitude
log_scale	ON	ON / OFF	Apply log(1 + |B|) — effective in Magnitude mode only
colormap	inferno	inferno / magma / viridis / plasma / hot / cool / gray / jet	Matplotlib colormap
width	768	256 – 4096	Output image width (px)
height	768	256 – 4096	Output image height (px)
show_axes	ON	ON / OFF	Show frequency axes, colorbar, and title

---

Output Modes

Magnitude mode (normalize = OFF)

Displays |B(f₁, f₂)|, optionally on a log scale. Bright regions indicate strong nonlinear coupling between frequencies $f_1$, $f_2$, and $f_1+f_2$.

Bicoherence mode (normalize = ON)

Displays the squared bicoherence:

$$b^2(f_1, f_2) = \frac{|B(f_1,f_2)|^2}{\mathbb{E}[|X(f_1)X(f_2)|^2]\cdot\mathbb{E}[|X(f_1+f_2)|^2]}$$

Values are normalized to [0, 1], making results comparable across signals of different amplitudes.

---

Computed Region

Only the triangular inner region is computed:

0 ≤ f₂ ≤ f₁,   f₁ + f₂ < Nyquist

This exploits the symmetry of the bispectrum and avoids redundant computation.

---

Installation

Option A — Clone into custom_nodes directly

cd /path/to/ComfyUI/custom_nodes
git clone https://github.com/bemoregt/ComfyUI_BiSpectrum.git

Option B — Symlink from a separate directory

git clone https://github.com/bemoregt/ComfyUI_BiSpectrum.git
ln -s /path/to/ComfyUI_BiSpectrum /path/to/ComfyUI/custom_nodes/ComfyUI_BiSpectrum

Restart ComfyUI after installation. No additional packages are required beyond those already used by ComfyUI (numpy, matplotlib, Pillow, torch).

---

Example Workflow

Load Audio ──► Audio to Bispectrum ──► Preview Image

Typical settings for music analysis:

• nfft = 512, hop_length = 256, win_length = 512
• normalize = ON (bicoherence), colormap = inferno

---

License

MIT