Load Audio

This is the entry point for basically everything else in the pack. Whatever you're doing - transcribing, extracting spectrograms, cutting and rejoining clips - it starts with getting an audio file into the graph as ComfyUI's AUDIO type, and this node is how you do that with a single file.

How it works

On first startup, the pack creates an audio_input folder (and a matching audio_output) inside ComfyUI's standard input/output directories. Drop your audio files in audio_input, and this node's file_name dropdown lists what's there - pick one, and it loads the waveform plus its sample rate as the pack's AUDIO type, which the README notes is compatible with ComfyUI's own official AUDIO implementation, so it interoperates cleanly with non-pack audio nodes too. There's nothing more sophisticated happening here than reading a file off disk - no resampling, no normalization, no format conversion. What you get out is exactly what's in the file.

The inputs and outputs that matter

• file_name - a dropdown populated from whatever's currently sitting in audio_input. It's the only input.
• audio (output) - the loaded waveform + sample rate, ready to feed into anything downstream that takes AUDIO.
• file_name (output) - the same filename you picked, passed through as a STRING. Handy for naming an output file to match the input later in the graph without hardcoding it twice.

How to install it

Via ComfyUI Manager: search ComfyUI-speech-dataset-toolkit, install, restart. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/kale4eat/ComfyUI-speech-dataset-toolkit
cd ComfyUI-speech-dataset-toolkit
pip install torchaudio --index-url https://download.pytorch.org/whl/cu121   # match your CUDA version
pip install -r requirements.txt

Restart ComfyUI. No model weights, no heavy dependency - this node is as lightweight as they come.

Common issues & troubleshooting

Dropdown is empty or missing a file you just added. ComfyUI scans the audio_input folder's contents when the page loads, not continuously. The README calls this out directly: if you drop a new file in while the app is already running, press F5 to reload the page before the dropdown will show it. This trips up almost everyone the first time.

Can't find the audio_input folder. It's created automatically on first startup of the pack, alongside audio_output, inside ComfyUI's input/output directories - not somewhere you need to create by hand. If it's genuinely missing, restart ComfyUI once with the pack installed and it should appear.

You wanted the official ComfyUI LoadAudio instead. That's a legitimate choice too - the README explicitly notes their AUDIO type is compatible with ComfyUI's own, so you can mix and match. Use whichever one's UI you prefer for picking the file; downstream nodes from this pack don't care which loader produced the AUDIO they're given.

Loading a whole folder of files at once. This node is single-file only. For a batch, use SDT_LoadAudios instead, which points at a subdirectory and returns a list of clips plus their filenames rather than one file at a time.