AUDIO Microphone Loader @ vrch.ai

The AUDIO Microphone Loader @ vrch.ai is the node that makes your microphone part of your workflow. Where the pack's Audio Recorder captures audio to record something, the Mic Loader streams live mic input as analysis data - volume, a waveform, a spectrum, per-band levels - so you can react to what you're saying or playing in realtime. It's the front half of every "make the visuals dance to my voice/music" setup, and it's the node people specifically hunt down for speech-to-text pipelines where the mic is the input and the text node is the output.

The neat part is the same trick the whole pack uses: the mic is captured in the browser via a web page (with all the permission prompts that implies), then streamed into the node. No audio routing software, no virtual cable, no OS-level config. Open the page, allow the mic, done.

How it works

The node opens a capture session on the device you pick, samples it, and emits several streams of numbers:

• WAVEFORM (FLOAT list) and SPECTRUM (FLOAT list) - the time-domain and frequency-domain views, ready for a visualizer or any math you want to do.
• VOLUME (FLOAT) - overall loudness, the simplest thing to react to.
• LOW_FREQ_VOLUME / MID_FREQ_VOLUME / HIGH_FREQ_VOLUME (FLOAT) - split-band levels, which is how you distinguish a bass drop from a cymbal crash.
• RAW_DATA (JSON) - the full analysis payload; this is what the pack's other analyzer nodes expect as input.
• IS_ACTIVE (BOOLEAN) - whether the stream is actually live, handy for gating logic so your workflow doesn't animate to silence.

The inputs you'll actually touch: device_id / name to pick a specific mic (leave both empty for the default), sensitivity (0–1) as a gain trim, sample_rate (16k / 24k / 48k), and frame_size (256 / 512 / 1024) which trades latency against detail - 256 feels snappier, 1024 gives smoother spectrum data. low_freq_max and mid_freq_max set where your low/mid/high band splits land. enable_preview shows you what the browser is hearing, which saves a lot of "why is this silent" debugging.

Installing it

Standard pack install:

cd ComfyUI/custom_nodes
git clone https://github.com/VrchStudio/comfyui-web-viewer
pip install -r comfyui-web-viewer/requirements.txt

Or install "ComfyUI Web Viewer" via ComfyUI Manager and restart. No models, no FFmpeg requirement for this node - that's the Recorder's burden, not this one's.

Where people get stuck

• No sound. The browser blocks mic access until you click Allow on the capture page's permission prompt, and if you denied it once you have to re-grant it in the browser's site settings. Also confirm the right device - on a laptop with a headset plugged in, "default" is a coin flip.
• A choppy waveform. Drop frame_size or the sample rate; the browser can only hand over frames as fast as the page can read them.
• The classic chain: Mic Loader → AUDIO Visualizer for the waveform/spectrum images, or Mic Loader → Frequency Band Analyzer → BPM Detector if you want rhythm-aware numbers feeding a sampler's parameters.