Podcast Transcript Analyzer

A Python command-line tool to analyze podcast transcripts from DOCX files and determine who talks more between Mark and Scott.

Features

• Word Count Analysis: Counts total words spoken by each speaker
• Speaking Time Estimation: Estimates speaking time based on word count and average speech rate
• Segment Analysis: Counts number of speaking segments per person
• Comprehensive Reports: Shows per-file and overall statistics
• Flexible Input: Analyze specific files or all DOCX files in directory
• Interactive D3.js Visualization: NY Times-style web interface with filtering, sorting, and animations
• Visualization Charts: Beautiful bar charts showing episode-by-episode and total speaking time
• CSV Export: Spreadsheet-ready data for further analysis

Installation

1. Make sure you have Python 3.7+ installed
2. Install the required dependencies:

pip install -r requirements.txt

Note: For visualizations, you'll need matplotlib and numpy (included in requirements.txt)

Usage

Quick Summary (Recommended):

python3 quick_summary.py

Shows a concise overview with key statistics and who talks more overall.

Detailed Analysis - All files:

python3 transcript_analyzer.py

Detailed Analysis - Specific files:

python3 transcript_analyzer.py "EP1 - Introduction.docx" "EP2 - Follow-up.docx"

Generate CSV Report:

python3 generate_csv_report.py

Creates a CSV file with detailed episode-by-episode statistics.

Generate Visual Charts:

python3 visualize_talk_time.py

Creates beautiful charts showing:

1. Episode-by-episode speaking time comparison (bar chart)
2. Total accumulated speaking time per speaker Saves as talk_time_analysis.png

Interactive Web Visualization:

# Generate data for interactive charts
python3 generate_json_data.py

# Open interactive_analysis.html in your browser
open interactive_analysis.html

Features NY Times-style interactive D3.js visualization with:

• Multiple view types (episodes, totals, percentages, timeline)
• Real-time filtering and sorting
• Smooth animations and hover tooltips
• Responsive design for all devices
• Client-side only (no server required)

Verbose output (shows individual segments):

python3 transcript_analyzer.py --verbose

How It Works

The analyzer uses several methods to identify speakers and content:

1. Speaker Detection: Looks for common patterns like:

   • Mark: content
   • Scott: content
   • [Mark] content
   • Mark - content

2. Word Counting: Counts words in each speaker's segments

3. Time Estimation: Uses average speaking rate (150 words/minute) to estimate speaking time

4. Aggregation: Combines results across all analyzed files

Output Format

The tool provides:

• Per-file breakdown showing words, estimated time, and segments for each speaker
• Overall totals with percentages
• Clear summary of who talks more by both word count and time

Your Podcast Analysis Results 🎉

Based on the analysis of 22 episodes:

🏆 WINNER: SCOTT TALKS MORE!

• Scott: 75,191 words (59.7%) | 8 hours 7 minutes speaking time
• Mark: 50,821 words (40.3%) | 5 hours 25 minutes speaking time

Episode Wins: Scott dominates 20 out of 22 episodes (90.9%)

Conclusion: Scott talks 19.3% more than Mark overall (+24,370 words, +2 hours 41 minutes)

Sample Output

PODCAST TRANSCRIPT ANALYSIS RESULTS
============================================================

File: EP1 - Introduction.docx
----------------------------------------
Mark :  150 words | 00:01:00 | 3 segments
Scott:  200 words | 00:01:20 | 4 segments

============================================================
OVERALL TOTALS
============================================================

Mark:
  Words:     150 ( 42.9%)
  Time:     00:01:00 ( 42.9%)
  Segments:     3

Scott:
  Words:     200 ( 57.1%)
  Time:     00:01:20 ( 57.1%)
  Segments:     4

Total words: 350
Total time:  00:02:20

============================================================
WHO TALKS MORE?
============================================================
By word count: Scott (+50 words)
By time:       Scott (+00:00:20)

Customization

You can modify the script to:

• Add recognition for additional speakers
• Adjust speaking rate estimation
• Change speaker detection patterns
• Add support for timestamps in transcripts

Notes

• The tool filters out files ending with :sec.endpointdlp
• Speaking time is estimated based on word count (150 words/minute average)
• If actual timestamps are present in transcripts, the tool can be enhanced to use them
• The analyzer is case-insensitive for speaker names