القائمة
Clean spoken videos by transcribing them, detecting pauses and filler words, and generating edited outputs with a conservative Korean-focused workflow.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
Create interactive HTML road-trip guides for multi-day driving itineraries. Use when an AI agent is asked to plan a road trip, build a travel itinerary page, make a fullscreen route map, estimate trip cost/time, compare restaurant or lodging alternatives, add POIs with photos and Google Maps links, or produce a self-contained HTML guide for any road trip.
Set up a new AI4PKM vault from scratch or migrate an existing vault to the latest template version. Detects current state, generates a plan, and applies changes with full backup safety.
고비 데스크탑 3.0 대화형 음성 온보딩 가이드
Create animated GIF slideshows from structured content. Use when the user asks to create a GIF summary, animated slide recap, GIF presentation, or visual summary GIF for events, meetups, presentations, or any multi-point content. Supports Korean text, dark/light themes, custom colors, progress indicators, and quote boxes. Ideal for social media sharing (Gobi Brain Updates, Slack, etc).
Convert DOCX files into markdown while preserving headings, lists, tables, metadata, and extracted images.
Convert EPUB files into a single well-formatted markdown document with extracted metadata, table of contents, and images.
| name | video-cleaning |
| description | Clean spoken videos by transcribing them, detecting pauses and filler words, and generating edited outputs with a conservative Korean-focused workflow. |
Automated video transcription and editing workflow that removes pauses and filler words from Korean videos using OpenAI Whisper API and MoviePy for frame-accurate cuts.
Use this skill when you need to:
Perfect for: Presentation recordings, lecture videos, podcast recordings, interview footage, or any speaking video that needs cleaning.
See the before/after comparison:
brew install ffmpegsudo apt install ffmpegpip install openaipip install moviepy (video editing with frame accuracy)OPENAI_API_KEY
This skill uses a two-step workflow:
transcribe_video.py){video_name} - transcript.json (complete API response){video_name} - transcript.md (formatted markdown){video_name} - word_timings.txt (simple text reference)edit_video_remove_pauses.py)This skill uses conservative editing to ensure safe, predictable results:
# Step 1: Transcribe your video
python transcribe_video.py "my_presentation.mp4"
# Step 2: Preview what will be removed
python edit_video_remove_pauses.py "my_presentation.mp4" --preview
# Step 3: Create edited video
python edit_video_remove_pauses.py "my_presentation.mp4"
# Basic transcription (Korean, default)
python transcribe_video.py "video.mp4"
# Custom output directory
python transcribe_video.py "video.mp4" --output-dir "transcripts"
# Keep extracted audio file
python transcribe_video.py "video.mp4" --keep-audio
# English transcription
python transcribe_video.py "video_en.mp4" --language "en"
Transcription outputs (auto-generated in same directory as video):
video - transcript.json (464 KB for 25-min video, ~2,800 words)video - transcript.md (formatted with timestamps)video - word_timings.txt (simple reference)# Preview mode (see what will be removed)
python edit_video_remove_pauses.py "video.mp4" --preview
# Basic editing (default settings)
python edit_video_remove_pauses.py "video.mp4"
# Custom pause threshold (remove pauses > 0.8 seconds)
python edit_video_remove_pauses.py "video.mp4" --pause-threshold 0.8
# Custom output path
python edit_video_remove_pauses.py "video.mp4" --output "cleaned_video.mp4"
# Adjust padding around cuts (default: 0.1 seconds)
python edit_video_remove_pauses.py "video.mp4" --padding 0.15
# Specify transcript location (if not auto-detected)
python edit_video_remove_pauses.py "video.mp4" --transcript "path/to/transcript.json"
# Skip filler word removal (only remove pauses)
python edit_video_remove_pauses.py "video.mp4" --no-fillers
# Save pauses data for chapter remapping (used by video-full-process)
python edit_video_remove_pauses.py "video.mp4" --output-pauses "video - pauses.json"
Editing outputs:
video - edited.mov (cleaned video)video - edited_edit_report.txt (detailed report)After editing, you'll receive a detailed report showing exactly what was removed:
============================================================
VIDEO EDIT REPORT (Conservative Mode)
============================================================
SUMMARY
------------------------------------------------------------
Original Duration: 00:25:21.000
Edited Duration: 00:23:16.690
Time Saved: 00:02:04.310 (8.2%)
Segments Kept: 67
PAUSES REMOVED
------------------------------------------------------------
Total Pauses: 43
Total Pause Time: 87.81 seconds
Top 10 Longest Pauses:
1. 5.64s at 00:07:51.500
2. 5.52s at 00:09:21.020
3. 3.54s at 00:00:11.520
...
FILLER WORDS REMOVED (Clear Fillers Only)
------------------------------------------------------------
Total Fillers: 28
Breakdown:
어 : 25 occurrences
아 : 3 occurrences
SAMPLE EDITS (First 5)
------------------------------------------------------------
1. Pause (3.54s) at 00:00:11.520
2. Pause (1.50s) at 00:00:29.400
3. Filler '어' (0.84s) at 00:05:32.140
...
Adjust based on your content:
# More aggressive pause removal (0.8 seconds)
python edit_video_remove_pauses.py "video.mp4" --pause-threshold 0.8
# Very conservative (only remove 2+ second pauses)
python edit_video_remove_pauses.py "video.mp4" --pause-threshold 2.0
Guidelines:
# Process all videos in a directory
for video in *.mp4; do
echo "Processing: $video"
python transcribe_video.py "$video"
python edit_video_remove_pauses.py "$video"
done
Always preview first when working with important content:
# Step 1: Transcribe
python transcribe_video.py "important_presentation.mp4"
# Step 2: Preview edits
python edit_video_remove_pauses.py "important_presentation.mp4" --preview
# Step 3: If satisfied, execute
python edit_video_remove_pauses.py "important_presentation.mp4"
The scripts follow consistent naming patterns:
Input:
presentation.mp4 (your original video)Transcription outputs:
presentation - transcript.json (complete data)presentation - transcript.md (formatted markdown)presentation - word_timings.txt (simple reference)Editing outputs:
presentation - edited.mov (cleaned video)presentation - edited_edit_report.txt (detailed report)The editing script uses MoviePy for frame-accurate cutting:
Segments shorter than 0.1 seconds are automatically filtered out:
Whisper API provides timestamps accurate to ~0.01 seconds:
두 파라미터는 컷 지점의 정확도를 조절합니다. Whisper 타임스탬프가 완벽하지 않기 때문에 버퍼가 필요합니다.
gantt
title 원본 오디오 타임라인
dateFormat X
axisFormat %s
section 원본
단어 A :a, 0, 2
휴지 (침묵) :crit, pause, 2, 5
단어 B :b, 5, 8
flowchart LR
subgraph 원본["원본 오디오"]
direction LR
A["🗣️ 단어 A<br/>0-2초"]
P["🔇 휴지<br/>2-5초"]
B["🗣️ 단어 B<br/>5-8초"]
A --> P --> B
end
flowchart TB
subgraph params["파라미터 작동 원리"]
direction TB
subgraph timeline["타임라인 (초)"]
direction LR
t0["0"] ~~~ t2["2"] ~~~ t5["5"] ~~~ t8["8"]
end
subgraph original["원본"]
direction LR
wordA["단어 A<br/>0~2초"]
pause["휴지<br/>2~5초"]
wordB["단어 B<br/>5~8초"]
wordA --> pause --> wordB
end
subgraph cuts["컷 포인트"]
direction LR
tail["◀── tail_buffer<br/>휴지 시작점을<br/>0.15초 뒤로 연장"]
pad["padding ──▶<br/>휴지 끝점에서<br/>0.1초 건너뜀"]
end
subgraph result["편집 결과"]
direction LR
keepA["✅ 유지: 0 ~ 2.15초<br/>(단어A + tail_buffer)"]
remove["❌ 제거: 2.15 ~ 5.1초"]
keepB["✅ 유지: 5.1 ~ 8초<br/>(padding 후 단어B)"]
keepA --> remove --> keepB
end
end
style pause fill:#ffcccc
style remove fill:#ffcccc
style keepA fill:#ccffcc
style keepB fill:#ccffcc
파라미터 요약:
| 파라미터 | 기본값 | 역할 | 값을 늘리면 |
|---|---|---|---|
--tail-buffer | 0.15초 | 휴지 시작 전 음성 보존 | 단어 끝부분 더 보존 |
--padding | 0.10초 | 휴지 끝난 후 건너뜀 | 더 많이 잘림 |
음성이 잘리는 경우:
--tail-buffer 증가 (예: 0.25)--padding 감소 (예: 0.05)Error: Transcript not found: video - transcript.json
Solution: Run transcription first or specify transcript location:
python transcribe_video.py "video.mp4"
# OR
python edit_video_remove_pauses.py "video.mp4" --transcript "path/to/transcript.json"
Solution: Transcript JSON is incomplete or corrupted. Re-run transcription:
python transcribe_video.py "video.mp4"
Error: ffmpeg not found
Solution: Install FFmpeg:
brew install ffmpegsudo apt install ffmpegError: OPENAI_API_KEY environment variable not set
Solution: Set your API key:
export OPENAI_API_KEY="sk-..." # Add to ~/.bashrc or ~/.zshrc
With MoviePy frame-accurate editing:
If you notice quality degradation:
preset=fastpreset=slowBased on typical Korean presentation videos:
Here's a complete example from start to finish:
# 1. You have a recorded presentation
ls
# presentation.mp4 (25:21 duration)
# 2. Transcribe the video
python transcribe_video.py "presentation.mp4"
# Output: Found 2,835 words
# Cost: $0.15
# Files created:
# - presentation - transcript.json
# - presentation - transcript.md
# - presentation - word_timings.txt
# 3. Preview what will be removed
python edit_video_remove_pauses.py "presentation.mp4" --preview
# Found 43 pauses > 1.0s
# Found 28 filler word instances
# Video will be split into 67 segments
# Time saved: 00:02:04 (8.2%)
# 4. Looks good! Create edited video
python edit_video_remove_pauses.py "presentation.mp4"
# Cutting video segments... (3 minutes)
# Concatenating segments... (1 minute)
# ✅ Success! Edited video saved to: presentation - edited.mov
# 5. Review the results
ls
# presentation.mp4 (original)
# presentation - edited.mov (cleaned, 23:17)
# presentation - edited_edit_report.txt (detailed report)
# presentation - transcript.json
# presentation - transcript.md
# presentation - word_timings.txt
python edit_video_remove_pauses.py "video.mp4" --preview
Review the report before committing to editing.
Never delete your original video until you've verified the edited version.
Try different --pause-threshold values on a test video to find what works best for your content.
Ensure audio is clear before transcription. Whisper works best with:
For multiple videos, process one completely first to verify settings, then batch the rest.
Total cost per video: Primarily OpenAI API fees (~$0.006/minute)
This skill can be combined with video-add-chapters for a complete video processing workflow. Use the video-full-process skill for an automated pipeline:
# Full processing with transcript reuse
python process_video.py "video.mp4" --language ko
The --output-pauses flag exports pause data in JSON format for chapter timestamp remapping:
{
"pauses": [
{"start": 45.2, "end": 48.5, "duration": 3.3},
{"start": 120.1, "end": 125.8, "duration": 5.7}
],
"total_pause_time": 87.5
}
This data enables accurate chapter remapping after pauses are removed.
ffmpeg -versionecho $OPENAI_API_KEY_Settings_/Skills/video-cleaning/
├── SKILL.md # This documentation
├── README.md # Quick reference
├── transcribe_video.py # Step 1: Transcription
└── edit_video_remove_pauses.py # Step 2: Editing
--no-fillers option┌─────────────────────────────────────────────────────────┐
│ VIDEO CLEANING WORKFLOW - QUICK REFERENCE │
├─────────────────────────────────────────────────────────┤
│ │
│ STEP 1: TRANSCRIBE │
│ python transcribe_video.py "video.mp4" │
│ → Creates: video - transcript.json │
│ │
│ STEP 2: PREVIEW │
│ python edit_video_remove_pauses.py "video.mp4" \ │
│ --preview │
│ → Shows: What will be removed │
│ │
│ STEP 3: EDIT │
│ python edit_video_remove_pauses.py "video.mp4" │
│ → Creates: video - edited.mov │
│ │
│ COMMON OPTIONS: │
│ --pause-threshold 0.8 Remove pauses > 0.8s │
│ --padding 0.15 Add 0.15s padding at cuts │
│ --output "clean.mp4" Custom output path │
│ --no-fillers Skip filler word removal │
│ │
│ WHAT GETS REMOVED: │
│ ✓ Pauses > 1.0 seconds │
│ ✓ Filler words: 어, 음, 아, 이, 오, 저 │
│ │
│ EXPECTED RESULTS: │
│ • 5-10% time reduction │
│ • 50-100 video segments │
│ • Natural-sounding output │
│ │
└─────────────────────────────────────────────────────────┘
Ready to clean your videos? Start with the basic workflow and adjust settings as needed. Remember: always preview first, and keep your originals safe!