video-delivery-coach

name	video-delivery-coach
description	Analyze YOUR video recordings before publishing. Evaluates voice (pace, pitch, volume), facial expressions (emotions, eye contact, smiles), and content (filler words, structure). Helps improve your Hinglish YouTube delivery over time.

Video Delivery Coach

Get better at video, video by video. This skill analyzes your recordings before you publish, identifying areas for improvement.

---

WHAT IT DOES

Analysis Type	Metrics	Tool Used
Voice	Speech rate (WPM), pitch variation, volume consistency	Librosa + Whisper
Facial	Emotion timeline, eye contact frequency, smile frequency	OpenCV + DeepFace + Mediapipe
Content	Transcription, filler words, structure	Faster-Whisper + Claude
Overall	5-dimension score (1-5 each, max 25)	Claude analysis

---

SCORING RUBRIC

Dimension	Score 1	Score 5
Content & Organization	Disorganized, unclear	Logical, well-structured
Delivery & Vocal Quality	Monotone, many fillers	Clear, varied, engaging
Body Language & Eye Contact	No eye contact, stiff	Direct gaze, natural movement
Audience Engagement	Boring, loses attention	Captivating, maintains interest
Language & Clarity	Grammar issues, unclear	Clear, impactful, professional

Total Score Interpretation:

• 5-9: Needs significant improvement
• 10-14: Developing skills
• 15-18: Competent speaker
• 19-22: Proficient speaker
• 23-25: Outstanding speaker

---

TRIGGERS

Use this skill when you say:

• "Analyze my video recording"
• "How was my delivery?"
• "Review my video before upload"
• "Check my presentation"
• "Coach my speaking"

---

USAGE

In Claude Code (Recommended)

"Analyze my video at /path/to/recording.mp4"

"Coach my delivery on the latest YouTube recording"

"What can I improve in this video?"

CLI Mode

# Basic analysis
python scripts/analyze_video.py --video "/path/to/video.mp4"

# Full analysis with all features
python scripts/analyze_video.py --video "/path/to/video.mp4" --full

# Voice only (faster)
python scripts/analyze_video.py --video "/path/to/video.mp4" --voice-only

# Save report
python scripts/analyze_video.py --video "/path/to/video.mp4" --output ~/reports/

---

OUTPUT FORMAT

Quick Summary

┌────────────────────────────────────────┐
│     VIDEO DELIVERY ANALYSIS            │
│     recording_2025_01_15.mp4           │
├────────────────────────────────────────┤
│  OVERALL SCORE: 18/25 (Competent)      │
│                                        │
│  Content & Organization:    4/5        │
│  Delivery & Vocal Quality:  3/5        │
│  Body Language & Eye Contact: 4/5      │
│  Audience Engagement:       4/5        │
│  Language & Clarity:        3/5        │
└────────────────────────────────────────┘

Detailed Report

# Video Delivery Analysis

**File:** recording_2025_01_15.mp4
**Duration:** 12:34
**Date:** 2025-01-15

---

## VOICE ANALYSIS

| Metric | Value | Target | Assessment |
|--------|-------|--------|------------|
| Speech Rate | 145 WPM | 120-160 | ✅ Good |
| Pitch Variation | 42.3 Hz | >30 Hz | ✅ Engaging |
| Volume Consistency | 0.08 | <0.15 | ✅ Steady |

**Filler Words Detected:**
- "um" - 8 times
- "you know" - 5 times
- "basically" - 3 times

**Recommendation:** Reduce "um" usage. Try pausing instead.

---

## FACIAL ANALYSIS

| Metric | Value | Assessment |
|--------|-------|------------|
| Eye Contact Frequency | 72% | ✅ Good |
| Smile Frequency | 35% | ⚠️ Could increase |

**Emotion Timeline:**
- 0:00-2:00: Neutral (intro)
- 2:00-8:00: Happy/Engaged (main content)
- 8:00-10:00: Serious (data presentation)
- 10:00-12:34: Happy (conclusion)

**Recommendation:** More smiles during technical sections.

---

## CONTENT ANALYSIS

**Strengths:**
- Clear opening hook
- Good use of clinical examples
- Strong call-to-action

**Areas for Improvement:**
- Could use more pauses after key points
- Consider adding more Hinglish transitions
- Section on side effects could be more structured

---

## OVERALL FEEDBACK

**What You Did Well:**
1. Excellent pace - not too fast, not too slow
2. Good eye contact with camera
3. Clinical examples were relatable

**What to Improve:**
1. Reduce filler words (especially "um")
2. Add more smiles during technical explanations
3. Pause after key statistics for emphasis

**Score: 18/25 - Competent Speaker**
You're delivering solid content with room for refinement.

---

HINGLISH-SPECIFIC ANALYSIS

This skill is calibrated for Hinglish content:

Feature	What It Checks
Code-switching	Natural Hindi ↔ English transitions
Pace adjustment	Slower for English technical terms
Cultural markers	Use of "ji", "beta", "aapko bata doon"
Engagement phrases	"Dekho", "Suniye", "Samjhe?"

---

COMPARING OVER TIME

Track your improvement across recordings:

┌─────────────────────────────────────────────────────┐
│  PROGRESS TRACKER (Last 5 Videos)                   │
├─────────────────────────────────────────────────────┤
│  Video           │ Score │ Main Improvement         │
│  ─────────────────────────────────────────────────  │
│  Jan 10          │ 15/25 │ Baseline                 │
│  Jan 15          │ 18/25 │ Better eye contact       │
│  Jan 20          │ 17/25 │ Fewer filler words       │
│  Jan 25          │ 19/25 │ More varied pace         │
│  Jan 30          │ 21/25 │ Natural Hinglish flow    │
└─────────────────────────────────────────────────────┘

---

INTEGRATION

With Your Workflow

Record Video → Analyze with video-delivery-coach → Fix issues → Re-record (optional) → Publish

Feeds Into:

• youtube-script-master - Script adjustments based on delivery feedback
• Personal improvement tracking

---

DEPENDENCIES

# Core (required)
pip install anthropic python-dotenv rich

# Voice analysis
pip install librosa moviepy faster-whisper

# Facial analysis (optional - for full analysis)
pip install opencv-python mediapipe deepface tf-keras

# Note: tf-keras is heavy (~500MB). Skip for voice-only mode.

---

API KEYS NEEDED

Key	Purpose	Status
ANTHROPIC_API_KEY	Final analysis and coaching	Already have

---

MODES

Voice-Only Mode (Lightweight)

python scripts/analyze_video.py --video file.mp4 --voice-only

• Requires: librosa, moviepy, faster-whisper
• Analyzes: Speech rate, pitch, volume, transcription, filler words
• Skip: Facial analysis (faster, lighter)

Full Mode (Comprehensive)

python scripts/analyze_video.py --video file.mp4 --full

• Requires: All dependencies including OpenCV, DeepFace, Mediapipe
• Analyzes: Everything including facial expressions
• Slower but complete

---

HOW CLAUDE SHOULD USE THIS SKILL

When user asks to analyze a video:

Step 1: Check if video file exists

import os
if not os.path.exists(video_path):
    print("Video file not found")
    return

Step 2: Run analysis

python scripts/analyze_video.py --video "/path/to/video.mp4"

Step 3: Present results

• Show quick summary first
• Offer detailed breakdown if requested
• Provide actionable recommendations

Step 4: Track progress

• Compare with previous analyses
• Note improvements
• Identify persistent issues

---

SAMPLE OUTPUT

=== VIDEO DELIVERY ANALYSIS ===
File: hinglish_statin_video.mp4
Duration: 15:23

VOICE METRICS:
├── Speech Rate: 138 WPM (Target: 120-160) ✅
├── Pitch Variation: 38.5 Hz ✅ Natural variation
└── Volume: Consistent ✅

FILLER WORDS:
├── "um": 12 occurrences
├── "basically": 8 occurrences
└── "you know": 5 occurrences

FACIAL METRICS:
├── Eye Contact: 68% ✅ Good
├── Smiles: 28% ⚠️ Below target (40%)
└── Dominant Emotion: Engaged

CONTENT SCORE:
├── Content & Organization: 4/5
├── Delivery & Vocal Quality: 3/5
├── Body Language: 4/5
├── Engagement: 4/5
└── Language & Clarity: 4/5

TOTAL: 19/25 (Proficient Speaker)

TOP 3 IMPROVEMENTS:
1. Replace "um" with pauses
2. Smile more during technical explanations
3. Slow down slightly when explaining statistics

HINGLISH NOTES:
✅ Natural code-switching
✅ Good use of "aapko batata hoon"
⚠️ Consider more "samjhe?" checks for engagement

---

NOTES

• Privacy: All analysis is local, video never uploaded anywhere
• Speed: Voice-only takes ~1 min, full analysis takes ~3-5 min
• File types: Supports MP4, MOV, AVI, MKV
• Duration: Works best with 5-30 minute videos

---

This skill helps you improve your delivery over time - not by judging, but by giving you objective data to work with.