SuperAPI AI Directory

soundcraft-voice-silence-gaps

General & other

View repository

Updated 11-04 23:02

About this skill

Silence region processing that preserves the timeline
  1. First diagnose the noise floor and the quietest speech; -40dB/100ms/10ms/20ms are only synthetic examples, not universal podcast presets. For real material, first try a short segment on a copy.
  2. The Audacity manual provides methods for judging threshold and duration, but its truncation deletes time; this SoundCraft workflow preserves the positions of subsequent speech, making it suitable for preserving synchronization.
  3. Check each new clip's start, length, source offset, and effective word start and end; do not look only at the clip count. For multi-track music or video, confirm synchronization positions separately.
  4. Output an explicit original range, and retain trailing silence even if the last segment ends early; do not treat the automatic project end as the total duration of the original material.

Before execution, read step-by-step parameters and invocation and checks. First read the real input, and replace all durations, tracks, and clip IDs according to the actual content. For existing projects, do not execute session.new, and do not overwrite the source session or original audio.

When embedded visuals are needed, hand off to the soundcraft-editor skill. Standalone installation:npx skills add full-aigc-skills/soundcraft-skills --skill soundcraft-editor. This batch of skills is provided as a pre-release with SoundCraft development; real material and other hosts still require separate acceptance testing.

The two valid signal segments retain their original time positions of 0.5s and 2.5s, with 10ms leading and 20ms trailing margins retained; other tracks remain unchanged; after exporting by the original 4-second range, compare the valid regions and the gap.

This method creates gaps; it is not Audacity Truncate Silence's time-deleting shortening, and it is not used to automatically compress total duration by pauses. Below the threshold there may be breath, tails, or music, which cannot be treated as nonexistent. It is not noise reduction or voice recognition.

Real-material listening, perceived loudness standards, and host visual synchronization must be verified separately, and must not be replaced by clip structure or a successful export.GitHub source and license。