Play Audio

The simplest node in the pack, and one of the more useful ones day to day: it plays a clip right there in the graph, no save step required. When you're iterating on a pipeline - tweaking VAD thresholds, checking a Demucs stem sounds right, listening to whether your resample introduced artifacts - dropping a SDT_SaveAudio node, running the graph, and then digging the file out of output/audio_output every single time is friction you don't need. This node skips all of it.

How it works

It's a terminal output node (is_output_node: true) with no output socket of its own. Wire an AUDIO clip into it, run the graph, and ComfyUI renders a player for that clip directly in the node's UI - the same pattern as the vanilla SaveAudio/PreviewAudio-style nodes, just scoped to this pack's audio type.

The inputs and outputs that matter

• audios - the only input, required. Note the plural naming (not audio) - worth a glance if you're wiring by variable name in a script rather than dragging the connection in the UI.

That's the entire schema. Nothing to configure, nothing it hands downstream - it's a listening point, not a processing step.

How to install it

ComfyUI Manager: search ComfyUI-speech-dataset-toolkit, install, restart. Manual clone if you'd rather:

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 for this node specifically - same torchaudio baseline as everything else in the pack.

Common issues & troubleshooting

Nothing plays, no player shows up. Make sure the graph actually executed up to this node - a PlayAudio sitting downstream of a node that errored or hasn't run yet just won't have anything to render. Check the queue/run status before assuming the node itself is broken.

Audio plays but sounds wrong compared to what you expect. This node just plays whatever AUDIO tensor it's handed - if something upstream (a resample, a Demucs stem split, a VAD-trimmed segment) already altered the clip, what you're hearing is that transformation, not a bug in playback. Drop a SDT_PlayAudio earlier in the chain to isolate where the change happened.

Running this in a headless/API context and expecting output. This node is built for the interactive editor UI - it renders a player widget in the browser. If you're driving the graph via the REST API rather than the web UI, there's no browser to render into, so don't build a pipeline that depends on this node's playback for anything programmatic; use it for manual iteration only.