Silence Audio

Not a trim, not a cut - this one mutes a window of a clip in place while leaving the total duration exactly as it was. That's a different job from SDT_TrimAudio/SDT_CutAudio, which change how long the clip is. Reach for this when you need to redact something - a mistake, a name, a stray noise - from the middle of a recording without shifting the timing of everything that comes after it.

How it works

It zeroes out the waveform samples between start_second and end_second, leaving every other sample untouched. Because nothing is removed - just silenced - the clip's total length, and any external timing that references it (an existing transcript, subtitle timestamps, an alignment file), stays valid after the operation. That's the entire reason this node exists as something distinct from trimming.

The inputs and outputs that matter

• audio - the clip to edit, required.
• start_second and end_second (both default 0, step 0.001) - the window to silence, in seconds down to millisecond precision.
• audio (output) - the same clip, same length, with that window zeroed out.

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

No extra dependencies beyond the pack's torchaudio baseline.

Common issues & troubleshooting

"I wanted to shorten the clip and it's still the same length." That's expected - this node never changes duration, by design. If you actually want to remove time from the clip rather than mute it in place, use SDT_TrimAudio or SDT_CutAudio instead.

Not sure the window is right. Sanity-check start_second/end_second against the clip's real duration before running blindly - SDT_AudioProperty gives you the sample rate to do the math, or just eyeball it first with SDT_PlotWaveForm.

Worried silencing will break an existing transcript's timing. It won't, and that's the point - because the clip's duration is unchanged, any timestamp-based alignment computed before you silenced a section stays valid afterward, which is exactly why you'd choose this node over cutting when you're working against an existing transcript or timed subtitle track.

Need to silence several separate windows in one clip. There's no batch/multi-range input here - one call handles one window. Chain multiple SDT_SilenceAudio nodes in sequence, one per window, feeding each one's audio output into the next node's audio input. Order doesn't matter as long as the windows don't overlap; if they do overlap, silencing is idempotent (re-zeroing an already-silent stretch changes nothing), so it's safe either way.