LLaSM Load Audio

Every multimodal pipeline needs a boring front door, and for the LLaSM pack that's this node. It does exactly one job: turn an audio file on your disk into the AUDIO format the LLaSM Interface node expects. It's the Load Image of the speech world - unexciting, necessary, and easy to take for granted.

How it works

The node's single input, audio, is a dropdown that lists every audio file sitting in your ComfyUI/input directory. It accepts .wav, .mp3, .ogg, .flac, .aiff, and .aif. Pick one, and librosa loads it and resamples to 16kHz - which happens to be exactly the sample rate Whisper's processor wants, so you never have to think about it. What comes out the AUDIO pin is a small bundle: the waveform plus its sample rate, ready for the Interface node.

The one thing to remember

That audio input is a dropdown, not an upload widget - the node deliberately disables the upload button. So your workflow is: drop files into ComfyUI/input, refresh the node, pick from the list. Files you add while the UI is already running won't show up until the node is rebuilt, so if something's missing, that's why.

Wiring it up

One input, one output. The AUDIO socket plugs straight into the LLaSM Interface node's audio input. Worth knowing: you're not locked into this node as the only source. Anything that emits ComfyUI's standard AUDIO type - VideoHelperSuite's audio extraction, for instance - plugs into the Interface node just as well. This node is simply the zero-dependency way to get there.

That's the whole job. It's a thin utility, and it's fine that it's a thin utility - the interesting part of the pack is what happens after the file gets in.