Plot WaveForm

The simplest node in the visualize category, and probably the one you'll reach for most. Amplitude over time, no transform, no parameters - just a picture of what your audio actually looks like. When a clip somewhere in a longer pipeline "seems wrong," this is the first thing to plug in before reaching for anything fancier like a spectrogram or pitch contour.

How it works

It plots the raw waveform amplitude across the clip's duration and returns the result as an image. That's the entire operation - there's no processing, filtering, or transform happening between input and plot.

The inputs and outputs that matter

• audio - the clip to visualize.
• title (default "Waveform") - cosmetic label.
• graph_image (output, IMAGE) - the rendered plot.

How to install it

Via ComfyUI Manager: search ComfyUI-speech-dataset-toolkit, install, restart. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/kale4eat/ComfyUI-speech-dataset-toolkit
cd ComfyUI-speech-dataset-toolkit
pip install torchaudio --index-url https://download.pytorch.org/whl/cu121
pip install -r requirements.txt

No extra dependencies beyond the pack baseline - this is about as lightweight as a node gets.

Common issues & troubleshooting

Flat line at zero. The clip is silent, or you're plotting something you didn't mean to - check whether SDT_SilenceAudio or SDT_MakeSilenceAudio touched this audio earlier in the graph, or whether a trim went out of the clip's actual bounds and returned nothing.

Peaks look flat-topped instead of rounded. That's clipping in the source recording - the waveform hit the format's amplitude ceiling and got chopped off. This node just shows it to you; the fix is a cleaner source recording, not anything downstream.

A sudden jump or discontinuity partway through the clip. Usually a bad splice - a concatenate or join node stitching two clips together without matching levels or a crossfade. Catching that here, as an ugly step in the waveform, is exactly how you'd spot it before it turns into an audible click or pop in a saved file.

When a waveform plot isn't enough. Amplitude over time tells you a lot, but not everything - it won't show you frequency content or intonation. If a clip looks fine here but sounds wrong, SDT_PlotSpecgram (frequency content) and SDT_PlotPitch (intonation contour) are the next two nodes to reach for before assuming there's nothing more to diagnose.

Habit worth building. Given how cheap this node is to run, it's worth wiring one in after any edit step - trim, silence, concatenate - while you're still building a pipeline, purely as a sanity check before you trust the output further downstream. Pull it back out once the graph is stable.