☁️Listen Audio

This is a push-to-talk microphone node. You hold a key down, it records; you let go, it stops and hands you the clip. It exists because comfyui_LLM_party isn't just about text - the README talks about building "the one-stop LLM + TTS + ComfyUI workflow" for things like streaming setups and voice assistants, and voice input has to come from somewhere. This is that somewhere.

The mechanism is deliberately simple: it doesn't listen continuously and it doesn't do voice-activity detection. It watches for a specific key, and as long as you're holding it, it's capturing from your microphone. That's a real design choice, not a limitation - continuous listening in a graph-based tool is a good way to accidentally record yourself thinking out loud, and a hold-to-talk pattern means you're always in control of exactly what gets sent onward, which matters a lot once whatever's downstream is an LLM that's about to act on what it hears.

The one input that matters. press_key is an enum with five choices - shift, space, ctrl, alt, tab - defaulting to shift. Pick whichever key doesn't collide with something else you're already using (space is an easy trap if you're also navigating the ComfyUI canvas with keyboard shortcuts open in the same window). That's the entire required config; there's no sample rate, duration limit, or device selector exposed on the node itself.

The two outputs. audio_path is a STRING pointing at where the recording got saved to disk, and audio is the actual AUDIO object other ComfyUI nodes expect on a socket. Which one you use depends on what's next in your graph - a node that wants a file path (say, something shelling out to a CLI transcription tool) takes audio_path; a node built around ComfyUI's native audio type takes audio directly. Having both means you're not stuck picking one convention over the other.

Installing it. This ships with the full comfyui_LLM_party pack - there's no separate audio-node install. Get it through ComfyUI Manager (search "comfyui_LLM_party") or cd ComfyUI/custom_nodes && git clone https://github.com/heshengtao/comfyui_LLM_party, run pip install -r requirements.txt from the project folder inside your ComfyUI Python environment, and restart. Audio capture is one of the few functions in this pack that needs real hardware access rather than just an API key, so it's also one of the few nodes here that genuinely doesn't make sense on a headless server or a remote/hosted execution environment - there has to be an actual microphone attached to the machine ComfyUI is running on, which rules out most cloud and serverless setups by default.

Where it actually goes wrong. If the node loads but recording produces nothing, the almost-always cause is that the machine running ComfyUI either has no default input device or the OS never granted the Python process microphone permission - check your system's audio input settings before assuming the node is broken. If the key-hold doesn't register at all, confirm the ComfyUI browser tab actually has focus when you press it; hotkey capture for a node like this generally can't fire if the window isn't active. And if the pack fails to import entirely rather than this one node misbehaving, that's the same general story as the rest of comfyui_LLM_party: it's a large requirements.txt covering audio, database, and API integrations all at once, real users have reported the whole node set failing to load after install, and the fix is almost always a dependency version conflict - worth trying the README's requirements_fixed.txt before digging further.