Audio Analyzer Path (Allergic)

ComfyUI got good at making sound before it got good at telling you what sound it's looking at. This node is the boring-but-useful answer to the second half: paste a file path to any audio file and it hands back the BPM, the musical key, the scale, and a plain AUDIO output you can route anywhere else in the graph. It's the "for minimalists" half of the two-node AllergicAudio pack by usrname0 - the one for people who type paths instead of picking files.

How it works

Under the hood it's a small, honest script. It loads the file with librosa, runs librosa's beat tracker to get a tempo, then computes a chroma (pitch-class) distribution with chroma_cqt and correlates it against the classic Krumhansl–Schmuckler key profiles - twelve rotations of the major and minor templates, whichever correlation wins decides the key. That's the textbook music-theory algorithm, not a learned model, so there's no model download and no GPU involved. You could reproduce the math in a notebook; this just drops it into a ComfyUI node.

The AUDIO output is built with PyAV and returned in the exact {"waveform": tensor, "sample_rate": int} shape ComfyUI's built-in LoadAudio produces, so anything that accepts an AUDIO input will take it. One honest caveat: PyAV isn't in the pack's declared requirements - only librosa is. It works because ComfyUI ships av anyway, but it's one of those quietly-undeclared dependencies you'd only notice in a stripped-down environment.

The input and outputs

The node is minimal by design. There's exactly one input:

• file_path (STRING) - a raw filesystem path in plain text. No picker, no drag-and-drop; paste it in and run.

The path gets run through the pack's sanitize_path helper, which strips surrounding quotes, preserves Windows drive letters, and drops the characters Windows forbids in filenames. Outputs:

• AUDIO - decoded audio as a ComfyUI AUDIO tensor, same format as LoadAudio
• bpm (INT) - rounded tempo
• key (STRING) and scale (STRING) - e.g. "G" and "minor"
• keyscale (COMBO) - the two combined as "G minor". It's deliberately typed as a COMBO so it'll plug into dropdown/combo inputs on other nodes that expect a value from a list.

Installing it

The whole pack installs one way regardless of which node you use. Easiest: ComfyUI Manager, search "Allergic Audio", install, restart - Manager handles the one declared dependency, librosa. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/usrname0/ComfyUI-AllergicAudio.git
cd ComfyUI-AllergicAudio
pip install .

On the Windows portable build, replace pip with ..\..\..\python_embeded\python.exe -m pip install .. Note librosa is a heavyweight - it drags in numba, llvmlite, scipy and friends, so the first install takes a few minutes and the first analyze run compiles numba JIT code and can stall for a moment. That's normal; the second run is fast.

Gotchas

The name Path is doing real work: this node reads any path the ComfyUI machine can reach, not just your input folder - handy for scripting or API-driven workflows. But the path must be valid from the server's point of view, not your browser's; if ComfyUI runs in Docker or on another box, that's the filesystem that matters.

One trap worth knowing: sanitize_path is Windows-flavored. It rewrites forward slashes to backslashes and collapses duplicates, which is fine on Windows but means a path like /home/you/loop.wav gets mangled into a literal backslash filename on Linux or macOS. If you're not on Windows, the Upload sibling in this same pack (file-picker UI, no path typing) is the safer route - or just accept that this node was tuned for the platform its sanitizer was written for.