deadspace_trimmer

Reads a transcript and removes silence, keeping only speech regions.

When to use

Use after generate_transcript to automatically cut out pauses, dead air, and long silences. This is a fast, deterministic alternative to the AI content cut for cases where you only want to remove silence (not make editorial decisions about content).

In the editor

Add it from the Library (Utility group), or with + add step and a search for deadspace_trimmer, after the step that produced your transcript.

Under Inputs, add transcript as a step input pointing at that transcript step. Two parameters shape the result: min_silence_duration (pauses shorter than this are left alone) and padding (a little audio kept either side of each kept region, so the cuts do not clip the start of words).

Tick Review gate if you want the run to stop and show you the cut before it goes any further. Approving keeps it, and rejecting or regenerating gives you another pass with different settings.

The run timeline shows one clip per stretch of speech, with the silences gone. Too many very short clips means it is cutting too eagerly: raise min_silence_duration. Long silences surviving means the opposite.

YAML Example

- id: clean_cut
  processor: deadspace_trimmer
  input:
    - id: transcript
      step: transcribe_main
  params:
    min_silence_duration: 0.5
    padding: 0.1
  output:
    timeline: main
  review: true
  cache: true

Inputs

Params

Output

EditDecisions: one edit decision per speech region.

Notes

• Supports review: true. Check the output to confirm it didn't accidentally cut important content.
• Increase min_silence_duration if it's cutting too aggressively (producing too many short clips). Decrease if large silences are surviving.
• padding prevents hard cuts at word boundaries: the small buffer around each region keeps natural breath sounds.