Load Audios

SDT_LoadAudio gets you one file at a time. This is the batch version: point it at a folder and it hands back every audio file inside as a list, ready to run through the same processing steps in one pass instead of wiring up a separate LoadAudio node per file. For anything that actually deserves the word "dataset" - a folder of recordings you're transcribing, resampling, or feeding into a spectrogram node en masse - this is the node you start from instead.

How it works

It reads a directory relative to the audio_input folder (the same folder SDT_LoadAudio reads single files from, created automatically on first startup) and loads every audio file it finds there, returning both the waveforms and their filenames as parallel lists. Anything downstream built around ComfyUI's list-processing behavior - a for-each style loop, or a node further down the pack's own list that natively accepts lists - can then run over the whole batch without you touching each file individually.

The inputs and outputs that matter

• dir - a dropdown of subdirectories under audio_input. The default . means "the audio_input folder itself," so if you haven't organized your files into subfolders, this works out of the box on whatever's sitting there.
• audios (output, list) - every loaded clip as an AUDIO list. This is the exact field name SDT_JoinAudio's input expects, and it's not a coincidence - that node is built to accept a batch straight from this one.
• file_names (output, list) - the matching filenames, parallel to audios, in case you need to preserve which output corresponds to which input file further down the graph.

How to install it

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

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
pip install -r requirements.txt

No extra dependencies beyond the pack's standard torchaudio baseline.

Common issues & troubleshooting

Dropdown shows . and nothing else. That's expected if you haven't created subfolders inside audio_input - . (the folder itself) is always available. Create a subdirectory under audio_input and drop files in there if you want to organize multiple batches separately and switch between them from this dropdown.

New files not showing up. Same F5 rule as SDT_LoadAudio - ComfyUI reads the folder listing on page load, not live. Reload the page after adding files while the server's already running.

Downstream node processed the batch, but you can't tell which output belongs to which file. Use the parallel file_names list - it's returned in the same order as audios, precisely so you can match results back to their source file rather than guessing from order alone.

Mixed sample rates or channel counts across the batch cause errors further down. This node doesn't normalize anything on load - it just reads whatever's on disk, so a folder with a mix of 16kHz mono and 44.1kHz stereo files will pass that mismatch straight through. Run the batch through SDT_ResampleAudio before anything that expects uniform audio, like SDT_JoinAudio or a batched spectrogram extraction.