auto-cut-silences

Auto-Cut Silences

Wrapper around auto-editor — detects silent segments based on an audio threshold and rebuilds the file with those segments removed. Re-encodes only the affected ranges; the rest is stream-copied where possible.

Procedure

1. Resolve auto-editor

Auto-editor is installed by deps-setup into the plugin's shared uv venv. Locate it via the venv path recorded in preferences.json:

DATA_DIR="${CLAUDE_USER_DATA:-${XDG_DATA_HOME:-$HOME/.local/share}/claude-plugins}/video-editing"
PREFS="$DATA_DIR/preferences.json"
VENV=$(jq -r '.python_venv // empty' "$PREFS" 2>/dev/null)

if [ -n "$VENV" ] && [ -x "$VENV/bin/auto-editor" ]; then
  AUTO_EDITOR="$VENV/bin/auto-editor"
elif command -v auto-editor >/dev/null; then
  AUTO_EDITOR="$(command -v auto-editor)"
else
  echo "auto-editor not found. Run deps-setup and select auto-editor."
  exit 1
fi

2. Inputs

Field	Default
Source	required (video or audio)
Threshold	4% (auto-editor default — silence below 4% of peak)
Margin	0.2sec (keep 0.2s of padding around speech)
Min cut	0.2sec (don't cut a silence shorter than this)
Output	sibling <basename>.tight.<ext> (or project's working/ inside an index project)

Auto-editor accepts these as --silent-threshold, --margin, and --min-clip-length / --min-cut-length.

3. Run

"$AUTO_EDITOR" "$SRC" \
  --silent-threshold "$THRESHOLD" \
  --margin "$MARGIN" \
  --output "$OUT"

For more aggressive trimming (e.g. talking-head where pauses are abundant), suggest --margin 0.1sec --silent-threshold 3%.

For a preview-only run (don't write a new file, just report what would be cut):

"$AUTO_EDITOR" "$SRC" --export timeline -o /tmp/auto-cut-timeline.json

The exported timeline is a JSON list of kept ranges — useful for showing the user what would be removed before committing.

4. Report

Print before/after duration and the cut ratio: