Blob To Audio Node

The Blob To Audio Node is how audio comes back into your graph. Your HTTP nodes get WAV bytes from a TTS, voice-clone, or audio-processing API, and this node decodes them into ComfyUI's AUDIO type - the {"waveform", "sample_rate"} dict that every audio node in ComfyUI speaks. Wire it into PreviewAudio to listen, or SaveAudio to keep the file. It's the return half of the audio round trip, mirroring Audio To Blob Node on the way out.

It's one of the new utility nodes in ComfyUI-HttpRequestNodes, ahkimkoo's fork of felixszeto's ComfyUI-RequestNodes - and one of the two (with Blob To Video Node) that give the fork its audio/video identity. The canonical workflow: Binary Post Request Node sends your WAV to a voice-clone service → response_bytes comes back → this node turns it into an AUDIO clip → PreviewAudio. Round trip, no files on disk, no intermediate steps.

How it works

One input, bytes (BYTES), one output, audio (AUDIO). The source uses Python's standard wave module - no third-party audio library needed here - and reads the header to get channels, sample width, and frame rate, then converts the raw samples to float. It handles the three common sample widths: 16-bit (the standard, and what Audio To Blob Node always writes), 32-bit, and 8-bit. Anything else throws a clear "Unsupported sample width" error rather than silently garbling your audio.

The output waveform lands in [batch, channels, samples] shape with the correct sample_rate, so it plugs into PreviewAudio, SaveAudio, and anything else that expects a native AUDIO. The one limitation worth knowing: it's WAV-only. If your API returns MP3 or FLAC bytes, this node won't decode them - the author kept it to the format this pack itself produces, so the in-pack round trip is guaranteed.

Install

ComfyUI Manager → search ComfyUI-HttpRequestNodes → Install, restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/ahkimkoo/ComfyUI-HttpRequestNodes
pip install soundfile imageio imageio-ffmpeg

Category: RequestNode/Utils. This node itself uses only the Python standard library - but install soundfile anyway, because the pack's audio encoders need it. No models, no keys, no downloads.