Skip to main content
Audio Denoise

Audio Denoise · screenshot coming soon

​
What this node does

Audio Denoise reduces broadband background noise in an audio track, chosen by method:
  • afftdn — FFT-based denoiser (FFmpeg afftdn), good for steady hiss and hum. Its noise-reduction amount scales with strength.
  • anlmdn — non-local means denoiser (FFmpeg anlmdn), a gentler, detail-preserving approach; also scaled by strength.
  • silenceremove — not a denoiser but a silence trimmer (FFmpeg silenceremove): it cuts sections quieter than a threshold, useful for removing dead air.
It takes a COMFYTV_AUDIO snapshot (or a COMFYTV_VIDEO, whose audio track is used) and outputs processed COMFYTV_AUDIO plus an fx_spec. It has a ▶ Run (FFmpeg does the work). If no source is wired, it emits just the fx_spec for chaining. To feed native ComfyUI AUDIO in or out, insert a Bridge — see bridges.md.

​
When to use it

  • Clean up constant fan or air-conditioner drone under a voice recording (afftdn).
  • Reduce tape/mic hiss while keeping the recording sounding natural (anlmdn).
  • Auto-trim long silences from a raw take before editing (silenceremove).

​
Parameters

​
method

Which processor to run: afftdn (default), anlmdn, or silenceremove.

​
strength

Denoise amount, range 0 to 1, default 0.3. For afftdn this maps to a noise-reduction figure (roughly 0–40); for anlmdn it maps to the filter’s smoothing strength. Higher removes more noise but risks a “watery” or hollow sound. Not used by silenceremove.

​
silence_db

Threshold below which audio counts as silence, in dB. Range -80 to -20, default -50. Lower = only very quiet passages are cut. silenceremove only.

​
min_silence_s

Minimum duration (seconds) a quiet stretch must last before it is trimmed. Range 0.1 to 5, default 0.5. silenceremove only.

​
keep_silence_s

How much silence (seconds) to leave in place around kept audio, so cuts don’t sound abrupt. Range 0 to 5, default 0.5. silenceremove only.

​
Outputs

​
Tips

  • Start strength low (0.2–0.3) and raise gradually; over-denoising leaves obvious artifacts.
  • silenceremove changes the track’s length — downstream nodes that assume the original duration (e.g. lip-sync or fixed-length video) should account for that.
  • For narrow, tonal buzz (like 50/60 Hz mains hum) the hum mode of Audio Repair is more surgical than broadband denoise.
  • Audio Repair — de-click, de-clip, wavelet denoise, and mains-hum removal for specific defects.
  • Audio EQ — a highpass band can remove low-frequency rumble without a full denoiser.
  • FX Chain — renders multiple fx_spec steps (including this one) in a single pass.