Save Audio With Sequential Numbering

SDT_SaveAudio takes a fixed file_name, which is fine for one clip but a problem the moment you're saving many - run the graph twice and you overwrite yourself, or you have to wire up a unique name for every single output by hand. This node exists for exactly the situation a dataset toolkit cares about: splitting a long recording into chunks (VAD segments, trim splits, whatever) and saving each one out with an automatically incrementing filename instead.

How it works

It writes to the same output/audio_output folder the pack sets up on first launch, same as SDT_SaveAudio, but instead of an explicit file_name field, it derives the filename automatically by appending an incrementing number onto filename_prefix. Successive saves land as distinct files rather than clobbering each other - the whole point of the node.

The inputs and outputs that matter

• audio - the clip to save, required.
• file_format - the same eight-format choice as SDT_SaveAudio: wav, mp3, flac, vorbis, sph, amb, amr-nb, gsm.
• filename_prefix (default "ComfyUI") - the base name the sequence number gets appended to.

No file_name field here - that's the deliberate difference from SDT_SaveAudio. No outputs either; this is a terminal node (is_output_node: true).

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
pip install -r requirements.txt

Nothing extra beyond the pack's torchaudio baseline - same as SDT_SaveAudio.

Common issues & troubleshooting

Can't find the saved files. Same as SDT_SaveAudio - they're written to output/audio_output, not the regular output/ root. Look for filename_prefix followed by a sequence number.

Ran the graph twice in separate sessions and worried about collisions. If avoiding any overwrite matters for your pipeline, check output/audio_output before a big batch run rather than assuming numbering always restarts at zero - where the sequence actually picks up from (an in-session counter vs. scanning existing files) isn't something to guess at when you're about to save hundreds of clips.

mp3/vorbis or the legacy formats (sph, amr-nb, gsm) fail to encode. Same caveat as SDT_SaveAudio - these route through backend codecs that aren't always present in a minimal torchaudio install. If wav and flac save fine but a lossy or legacy format doesn't, it's a missing codec on your backend, not this node.

This is the node to reach for in a loop. If you're already looping over VAD chunks or trim splits and saving each one, this node is built for exactly that - no per-iteration naming logic required on your end.