🕵️‍♂️ AIIA Audio Info

Every ComfyUI audio user eventually hits the same moment of doubt: "is this file actually 24kHz, or did some node quietly resample it?" This node is the answer to that question. It inspects an AUDIO tensor and reports what's really in it - sample rate, duration, channel count, and the tensor shape. It's a diagnostic node, not a processing node: it changes nothing, tells you everything. In a pack this dense with TTS and voice-conversion models that each have opinions about sample rates, that's more useful than it sounds.

Why you'll actually reach for it

Sample-rate mismatches are the silent killer in audio graphs. A node trained on 24kHz audio fed 44.1kHz input doesn't error - it just produces subtly wrong output, or a pitched-up voice, or silence at the far end. When a talking-head video comes out with the mouth out of sync, or a voice-conversion result sounds a half-step higher than the source, the first thing you want to check isn't the model - it's whether the audio going in matches what the model expects. This node makes that a two-second check instead of a guessing game.

The inputs and outputs

One input: audio. Zero config. Five outputs, all passive:

• info_text (STRING) - a readable report of everything below, ready to drop into a text display node.
• sample_rate (INT) - the actual rate, e.g. 24000 or 44100.
• duration_seconds (FLOAT) - total length in seconds.
• channels (INT) - mono (1) or stereo (2).
• shape_info (STRING) - the raw tensor shape, e.g. batch × samples × channels. This is the "trust, but verify" output for when a downstream node cares about layout, not just rate.

Wire the info_text into a text preview and you get a live readout; or branch the numeric outputs into a compare/debug node if you're building automation around format checks.

Where it fits

Stick it right before any model node whose input format you're unsure about - before a voice-conversion target audio, before a TTS reference, before a talking-head drive track. If the number doesn't match what the README says the model wants, that's your bug, found in seconds instead of after a 20-minute render. It's also the honest answer to "why does my output sound weird" workflows: run the source through this node and the output through this node, and compare. That single habit will fix more audio pipeline confusion than any other debugging move.

The honest take

This is a plain utility - nothing fancy, no AI, no models to download. It's the kind of node you don't appreciate until a mystery bites you, at which point you're annoyed it wasn't already in your default graph. One small complaint: it won't tell you the file's original format, only the in-memory tensor, so if something upstream already resampled, that's all you'll see. Which, in practice, is still the answer you need.