silence-cutter

name	silence-cutter
description	Cleans up videos for editing workflows. Supports silence removal, transcript-driven filler/repetition cleanup, and hybrid mode. Exports an edit-friendly Premiere XML timeline with visible cut boundaries and optional markers, plus an optional cut MP4.
version	2.0.0
metadata	{"hermes":{"tags":["video","editing","ffmpeg","premiere","transcript","content-creation"],"category":"media","requires":["ffmpeg","ffprobe","python3"]}}

Silence Cutter

Use this skill when the user wants a rough cut prepared for Adobe Premiere Pro. It can:

• remove silence
• remove spoken filler words like uh, um, you know, i mean
• remove obvious repeated words / short repeated phrases
• export a Premiere-importable XML timeline with editable cut boundaries
• optionally export a cut MP4

The Premiere timeline stays editable. Editors can drag cuts earlier/later, extend clips, or remove cuts manually.

SKILL_DIR is the directory containing this SKILL.md file.

When To Use

Trigger this skill when the user says things like:

• "cut the silence from this video"
• "remove dead air"
• "clean up my talking head video"
• "remove um and uh"
• "clean repeated words from my English video"
• "make me a Premiere timeline from this raw recording"

Prerequisites

ffmpeg -version
ffprobe -version
python3 --version

For speech-clean and hybrid modes, install the transcript dependency:

pip install faster-whisper

Modes

• silence: remove silent gaps only
• speech-clean: remove spoken filler words and obvious repeated words/phrases
• hybrid: run speech cleanup and silence cleanup together

Core Commands

# Silence only
python3 SKILL_DIR/silence_cutter.py /path/to/video.mp4 --mode silence

# Speech cleanup only (English-first workflow)
python3 SKILL_DIR/silence_cutter.py /path/to/video.mp4 --mode speech-clean --language en

# Both together
python3 SKILL_DIR/silence_cutter.py /path/to/video.mp4 --mode hybrid --language en

# Premiere XML only
python3 SKILL_DIR/silence_cutter.py /path/to/video.mp4 --mode hybrid --xml-only

# More conservative speech cleanup
python3 SKILL_DIR/silence_cutter.py /path/to/video.mp4 --mode speech-clean --aggressiveness low

# Custom filler list
python3 SKILL_DIR/silence_cutter.py /path/to/video.mp4 \
  --mode speech-clean \
  --filler-words "uh,um,erm,like,you know,i mean"

Premiere Handoff

python3 SKILL_DIR/premiere_bridge.py /path/to/video.mp4 --mode hybrid --language en

This generates the Premiere XML and asks macOS to open it with Adobe Premiere Pro.

How To Report Results

After running, tell the user:

• which mode was used
• how many cut segments were found
• how much time was removed
• where the XML and MP4 were saved
• that the Premiere timeline is still editable
• that markers may show where silence/filler/repetition cuts happened

Premiere Import Instructions

Tell the user:

To import into Adobe Premiere Pro:
1. Open Premiere Pro
2. File -> Import
3. Select the generated _timeline.xml file
4. Premiere creates a new sequence with all cuts applied
5. Clip boundaries remain editable in the timeline
6. If markers were included, use them to review aggressive cuts quickly

Important Parameters

Parameter	Description
`--mode silence	speech-clean
--language en	Transcript language for speech cleanup
--speech-model base	faster-whisper model size
`--aggressiveness low	medium
--filler-words	Custom comma-separated filler list
--repeat-gap	Max gap between repeated words/phrases
--padding	Handle around cut edges
--no-markers	Export XML without cut markers
--xml-only	Export XML only
--mp4-only	Export MP4 only

Output Files

File	Description
{name}_timeline.xml	Silence mode Premiere timeline
{name}_speech_clean_timeline.xml	Speech-clean Premiere timeline
{name}_hybrid_timeline.xml	Hybrid Premiere timeline
{name}_cut.mp4	Silence mode cut video
{name}_speech_clean_cut.mp4	Speech-clean cut video
{name}_hybrid_cut.mp4	Hybrid cut video

Pitfalls

• speech-clean needs faster-whisper; without it the command fails with an install hint.
• The speech cleanup workflow is tuned for English first. Other languages need a custom filler list and should be reviewed manually.
• Repetition cleanup is heuristic-based. Review aggressive cuts in Premiere using clip boundaries and markers.
• If the XML feels too aggressive, rerun with --aggressiveness low.
• If silence cuts are too strong, raise --min-silence or use a less aggressive threshold.

Verification

1. Run the script and confirm the report looks reasonable.
2. Import the generated XML into Premiere.
3. Check that cuts appear as separate clip boundaries in the new sequence.
4. Inspect markers for silence / filler / repetition removal points.
5. Extend or trim clips manually inside Premiere where needed.

Example Session

User: "Clean up my English talking head video for Premiere. Remove ums and silences."

Hermes runs:
python3 SKILL_DIR/silence_cutter.py /Users/me/Desktop/raw.mp4 --mode hybrid --language en --xml-only

Result summary:
- Mode: hybrid
- Removed silence + filler/repetition segments
- Saved Premiere XML to /Users/me/Desktop/raw_hybrid_timeline.xml
- Timeline remains editable inside Premiere