Nodes/ComfyUI-SAM-Audio/SAM-Audio Text Separate
ComfyUI Node

SAM-Audio Text Separate

Separates a described sound from audio. Short lowercase noun or verb phrases such as 'man speaking' or 'dog barking' best match SAM-Audio training.

By ethanfel·Created 2 days ago·Updated 2 days ago· 3
SAM-Audio Text Separate
  • pipeline
  • audio
  • target
  • residual
descriptionman speaking
predict_spansfalse
seed0
inference_steps32
chunk_duration10.0
chunk_overlap1.0
Categoryaudio/SAM-Audio

Inputs (8)

NameTypeDefaultDescription
pipelineSAM_AUDIO_PIPELINE
audioAUDIO
descriptionSTRINGman speakingA concise lowercase description of the sound to isolate.
predict_spansBOOLEANfalseAsk supported models to locate non-ambient sound events before separation. Uses additional memory and time.
seedINT00–18446744073709550000Controls SAM-Audio's initial noise for reproducible separation.
inference_stepsINT322–128Number of midpoint function evaluations. Higher values are slower and may improve quality.
chunk_durationFLOAT10.00–3600Seconds processed per pass. Use 0 to process the entire clip at once.
chunk_overlapFLOAT1.00–60Seconds shared by adjacent chunks for a smooth crossfade.

Outputs (2)

NameTypeDescription
targetAUDIO
residualAUDIO