Audio math

Audio in ComfyUI is a real thing these days, but the stock nodes mostly give you load, save, and maybe a volume control. Audio math from More Math is a full per-sample expression evaluator over your waveforms: mix tracks with a formula, apply an envelope that follows the sample position, build a fade that's a mathematical curve instead of a fixed ramp.

How it works

The node evaluates your expression for every sample of every channel. What makes it worth your time is the audio-specific variables: S is the current sample index, T is the total sample count, R is the sample rate, C is the current channel, N is the channel count, B the batch index. That means a fade-in is literally I0 * (S/T), a pan is I0 * (C - (N-1)/2), and a slow tremolo is I0 * (0.5 + 0.5*sin(2*pi*5*S/R)) - no dedicated modulation nodes needed. All the shared functions apply too: clamp, smoothstep, abs, exp, plus FFT (fft/ifft) for frequency-domain experiments.

The default expression is the pack's a*(1-w)+b*w-style lerp - in the autogrow dialect that's I0*(1-F0)+I1*F0, a crossfade between audio V0 and V1 by float F0.

Inputs and outputs

V is the autogrow list of audio inputs (V0, V1, ...), F the floats, Expression the formula. length_mismatch (error default, plus tile and pad) decides what happens when inputs have different sample counts - the tooltip spells it out: tile repeats the shorter, pad treats missing samples as zero. batching controls frames per pass.

Outputs are AUDIO (a list) and STACK. The audio feeds the usual audio sinks - a save node, or the pack's own spectrogram converter.

Installing it

Standard for the pack:

cd ComfyUI/custom_nodes
git clone https://github.com/mcDandy/more_math
cd more_math
pip install -r requirements.txt

Restart ComfyUI, or grab "More math" from ComfyUI Manager. Dependency is antlr4-python3-runtime on top of torch. No model downloads.

Gotchas

Per-sample evaluation of a long clip is a lot of iterations; keep heavy loops out of the expression and let the vectorized functions do the work. Watch length_mismatch - mixing a 44.1kHz and 48kHz clip hits the default error immediately, and no amount of tile fixes a fundamentally different sample rate. And amplitudes: audio isn't normalized, so I0*2 genuinely doubles the volume and can clip on playback. The pack is new and solo-maintained, so the audio path is early-adopter territory - but for anyone doing generative sound design or sample math, it's the most capable node of its kind in the ecosystem right now.