Audio Information

ComfyUI audio is annoyingly opaque. A wire shows you an "AUDIO" label, but nothing about how many clips are riding on it, what sample rate it's running at, or whether it's actually silence that's breaking your workflow. That's what Audio Information (class SET_AudioInfo) is for. It's the "show me the receipts" node of the Audio Batch pack, and it costs you nothing but a tiny slice of your graph.

What it does

Feed any AUDIO value in and it hands you back a stack of facts. The useful ones:

• batch_size - how many audio items are in the batch. 1 means a single clip; anything higher means someone upstream combined several. This is the thing that answers "why is my video playing 6 sounds at once."
• sample_rate - the frequency in Hz (44100, 48000, 16000…). The single most common source of audio weirdness in ComfyUI is two clips at different rates being treated as if they match.
• channels - 1 for mono, 2 for stereo. Good for catching the moment a node quietly collapsed your stereo into fake mono.
• num_samples - total sample count. Duration in seconds is just num_samples / sample_rate, which the README helpfully spells out.

The three tensor outputs - mean, std, and peak - are per-item stats (one value per item in the batch). peak is the one you'll actually reach for: it's the loudest sample in each clip, which is exactly what SET_AudioNormalize reports back as its original_peak_level output if you ever need to undo a normalization. mean and std are the "is this clip secretly empty or blown out" debuggers - a near-zero mean with a tiny std is silence.

There's also audio_bypass, which just echoes the input straight through. That's a small genius move: you can drop this node in the middle of a chain to peek at the signal without having to rewire anything. The node sits in the audio/conversion category, though "inspection" would be a fairer label.

How it fits

This is a debugging node, not a pipeline node. You hang it on a wire when something sounds wrong: voice too quiet, two clips stacked instead of one, stereo where mono should be. The outputs are just INT and TORCH_TENSOR, so you can wire them into any text-display or debug node you already have. The pack ships example workflows that use it to show you exactly what a resample or a channel conversion did to your audio.

Install

It's part of the Audio Batch pack by set-soft (Salvador E. Tropea), which installs from ComfyUI Manager by searching "Audio Batch", or manually:

cd ComfyUI/custom_nodes/
git clone https://github.com/set-soft/ComfyUI-AudioBatch
pip install seconohe

Then restart ComfyUI. No models to download, no heavy deps - the pack only needs torch, torchaudio and numpy, which a stock ComfyUI already has, plus the tiny seconohe helper library.

Gotchas

The one real trap: num_samples changes with sample rate, so don't compare durations across clips at different rates without accounting for that. And if peak reads exactly 0.0, that clip is digital silence - stop debugging your downstream nodes and fix the source. For verbose logging of what each node in the pack is doing, set the AUDIOBATCH_NODES_DEBUG=1 environment variable before launching ComfyUI.