agentsclimarketplace

Remove pauses from track

Skill stefanwebb/premiere-ai-skills/skills/remove-pauses-from-track

Use when asked to remove pauses, silences, or dead air from a specific track in a Premiere Pro sequence, e.g. "remove the pauses from the narration track" or "cut the silences out of audio track 2 (and keep the camera video in sync)". Also use when a linked video track must stay frame-aligned with the audio track being cleaned up.From its SKILL.md

Install
npx -y skills add stefanwebb/premiere-ai-skills --skill remove-pauses-from-track

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 4 stars4 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

5.3 KB, as published. Nobody here has run it

/remove-pauses-from-track

Detects and ripple-deletes silent pauses from one audio track in a Premiere Pro sequence — and, if that audio is paired with a video track (e.g. narration synced to camera footage), applies the exact same cuts to that video track too, so the two stay in sync. Destructive: verify on a duplicate/throwaway sequence before running on real footage — there is no automated undo-grouping for the batch of cuts (each is its own undo step in Premiere).

Before running

Identify from context (or ask if ambiguous):

  • The sequence name (or confirm the active sequence is the right one).
  • The audio track index (0-based, e.g. 0 = Audio 1) to clean up.
  • Whether that audio is linked to a video track, and if so, its index (0-based, e.g. 0 = Video 1). There's no auto-detection of "linked" tracks — state it explicitly, or omit if the audio track has no corresponding video (e.g. a pure narration/VO track).

Call premiere-cli get-project-info to confirm the sequence exists (in the currently active project) and read its frameRate — needed for step 2 below.

Running the pipeline

  1. Export the track's audio (always the entire sequence — no --start-seconds/--end-seconds — so the exported file's time 0 is exactly the sequence's time 0, with no offset math needed later):

    premiere-cli extract-audio-track --sequence-name "<name>" \
      --audio-track-index <N> --format wav --output /tmp/<name>-track<N>.wav
    
  2. Detect pauses, passing the sequence's actual fps (from get-project-info above) so the cut list's frame quantization matches what remove-track-intervals will independently re-derive from the sequence itself — these must agree for the frame numbers to mean the same thing on both sides:

    remove-pauses /tmp/<name>-track<N>.wav --aggressiveness 1.0 \
      --min-pause 10 --fps <actual fps> -o /tmp/<name>-track<N>.cuts.txt
    

    (--aggressiveness 1.0 cuts tight to both edges of each pause; --min-pause 10 cuts pauses as short as 10ms — these are this skill's defaults, not remove-pauses's own CLI defaults, which are more conservative.)

    By default every VAD-confirmed silence is a cut candidate — no ANTHROPIC_API_KEY required. Add --phrase-boundaries to instead gate most cuts to word-gaps Claude judges to be phrase or sentence boundaries, so a mid-thought hesitation isn't clipped (this requires ANTHROPIC_API_KEY). Either way, any silence of 1 s or longer is cut wherever it falls — past a point a pause is not hesitation, it is dead air. Change that threshold with --always-cut <seconds>, or --always-cut 0 to cut nothing except word-gaps.

  3. Apply the cuts:

    premiere-cli remove-track-intervals --sequence-name "<name>" \
      --audio-track-index <N> [--video-track-index <M> ...] \
      --intervals-file /tmp/<name>-track<N>.cuts.txt
    

    A "could not reach the Premiere Bridge panel on port 47823 ... timed out" error here does not necessarily mean the cuts failed — applying a large batch of ripple deletes can leave the panel unresponsive to new requests until it finishes. Don't assume failure; move on to step 4 to check whether the cuts actually landed before retrying or reporting an error.

  4. Check for dead air left behind. Extract the resulting track and look for silence the pass should have removed:

    premiere-cli extract-audio-track --sequence-name "<name>" \
      --audio-track-index <N> --format wav --output /tmp/<name>-after.wav
    

    Then, in the premiere-pro env:

    from premiere_ai.remove_pauses import _detect_silence
    silence, duration = _detect_silence("/tmp/<name>-after.wav")
    print(sum(b - a for a, b in silence), "s of silence in", duration, "s")
    

    A little is expected — cuts are frame-quantised and every kept pause at a boundary counts. Seconds of it, or any single stretch over ~2 s, means something is being missed; report it rather than leaving it.

  5. Report the result in plain language — how many intervals were applied, how many segments removed, and any warnings from the result (a track where a segment couldn't be removed is surfaced, not silently dropped — the rest of the cuts still succeed).

  6. Clean up the temporary files (.wav, .words.json, .txt, .cuts.txt) created along the way.

Notes

  • remove-pauses only needs ANTHROPIC_API_KEY if --phrase-boundaries is passed (its Claude phrase/sentence-boundary detection step) — if it's requested but the key is missing, remove-pauses itself reports a clear error; nothing special needs to be done here.
  • Full flag reference for either underlying command: premiere-cli extract-audio-track --help-equivalent is the extract-audio-track section of the premiere-cli skill; remove-pauses --help for the pause-detection flags; remove-track-intervals section of the premiere-cli skill for the apply step.

Keep looking

Skills are one crate of 610,045. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.