Zoom2GoogleTranscript

Automatically transcribe Zoom mp4 videos to Google Docs using local Whisper AI with intelligent calendar integration and speaker identification. Zero ongoing costs - all processing happens on your machine.

✨ Key Features

• 🎥 Batch Processing - Transcribe multiple videos automatically
• 📅 Calendar Integration - Automatically fetches meeting titles and attendees from Google Calendar
• 👥 Speaker Identification - Uses calendar attendees for accurate speaker names
• 💰 Zero Cost - Uses local Whisper AI (no API charges)
• 📝 Google Meet Format - Creates properly formatted transcripts matching Google Meet style
• 🔄 Progress Tracking - Rich terminal UI with progress bars
• 🎯 Model Selection - Choose speed vs accuracy trade-off
• 🤖 AI Speaker Diarization - Optional advanced speaker detection with pyannote.audio

🚀 Quick Start

Prerequisites

1. Python 3.9+

2. ffmpeg - Required for audio processing

   # macOS
   brew install ffmpeg
   
   # Ubuntu/Debian
   sudo apt install ffmpeg

3. Google Cloud Project - For API access

   • Create project at https://console.cloud.google.com
   • Enable Google Docs API, Drive API, and Calendar API
   • Download OAuth 2.0 credentials as credentials.json

Installation

# Clone the repository
git clone https://github.com/stephenhsklarew/Zoom2GoogleTranscript.git
cd Zoom2GoogleTranscript

# Install dependencies
pip install -r requirements.txt

# Authenticate with Google
python authenticate.py

Basic Usage

# Transcribe all videos in a folder
python video_transcriber.py /path/to/zoom/recordings

# Use a specific model
python video_transcriber.py /path/to/zoom/recordings --model medium

# Specify credentials
python video_transcriber.py /path/to/zoom/recordings --credentials token_video.pickle

📋 Output Format

The tool creates Google Docs transcripts in Google Meet format:

Dec 9, 2024
Steve/Karan/Stephen - Transcript
Attendees: karan.apatel, stephen.sklarew, steve.burden
00:00:00

karan.apatel: Hey everyone, thanks for joining...
stephen.sklarew: Great to be here. Let's discuss...
steve.burden: I'll start with the quarterly results...

How It Works

1. Extracts date/time from Zoom folder names (format: YYYY-MM-DD HH.MM.SS Meeting Name)
2. Queries Google Calendar for matching events (±30 minute window)
3. Extracts meeting details - title and attendee list
4. Transcribes audio using Whisper AI
5. Maps speakers to calendar attendees
6. Creates formatted Google Doc with proper attribution

🎤 Speaker Identification

Method 1: Calendar-Based (Default)

Uses pause detection (>2 seconds) combined with calendar attendee names:

• ✅ Zero setup required
• ✅ Works immediately with calendar integration
• ✅ Good for 2-3 person conversations
• ⚠️ Less accurate for complex multi-speaker scenarios

Method 2: AI-Powered Diarization (Optional)

For advanced speaker detection with pyannote.audio:

1. Get Hugging Face Token:

   • Create account at https://huggingface.co
   • Accept agreement at https://huggingface.co/pyannote/speaker-diarization-3.1
   • Get token from https://huggingface.co/settings/tokens

2. Use with token:

   # Via environment variable (recommended)
   export HF_TOKEN=hf_your_token_here
   python video_transcriber.py /path/to/videos
   
   # Or via command line
   python video_transcriber.py /path/to/videos --hf-token hf_your_token_here

Benefits of AI Diarization:

• 🎯 More accurate speaker detection
• 👥 Better for 3+ person meetings
• 🔊 Analyzes voice characteristics, not just pauses
• ✅ Still free (runs locally)

🎛️ Command Line Options

python video_transcriber.py <video_folder> [OPTIONS]

Required:
  video_folder              Path to folder containing Zoom recordings

Optional:
  --model MODEL            Whisper model: tiny, base, small, medium, large
                           (default: base)

  --no-recursive           Don't search subdirectories

  --folder-id ID           Google Drive folder ID to save documents

  --credentials PATH       Path to Google credentials file
                           (default: token_video.pickle)

  --hf-token TOKEN         Hugging Face token for speaker diarization
                           (can also use HF_TOKEN environment variable)

  --since DATE             Only process videos modified after this date
                           Format: YYYY-MM-DD or YYYY-MM-DD HH:MM:SS
                           Examples: 2024-12-01 or "2024-12-01 14:30:00"

📊 Model Comparison

Model	Speed	Accuracy	RAM	Download Size
tiny	~32x realtime	Lowest	1GB	~75MB
base	~16x realtime	Good ✅	1GB	~140MB
small	~6x realtime	Better	2GB	~460MB
medium	~2x realtime	High	5GB	~1.5GB
large	~1x realtime	Best	10GB	~3GB

Recommendation: Start with base model for speed/quality balance.

Real-World Performance

MacBook Pro M1 (CPU only):

• 30 min video with base model: ~2 minutes
• 30 min video with medium model: ~4 minutes

🔧 Setup Details

1. Google Cloud Setup

1. Go to https://console.cloud.google.com
2. Create a new project
3. Enable these APIs:
   • Google Docs API
   • Google Drive API
   • Google Calendar API (v3)
4. Create OAuth 2.0 credentials:
   • Application type: "Desktop app"
   • Download as credentials.json
   • Place in project directory

2. Authentication

Run the authentication script once:

python authenticate.py

This will:

• Open your browser for Google OAuth
• Request permissions for Docs, Drive, and Calendar
• Save credentials to token_video.pickle

The token is reused for all future transcriptions.

3. Zoom Recording Structure

For calendar integration to work, organize recordings in Zoom's default format:

Zoom/
├── 2024-12-09 10.31.25 Steve_Karan_Stephen/
│   └── video1487928882.mp4
├── 2024-12-02 15.00.18 Diane_Stephen Weekly 1_1/
│   └── video1683623283.mp4
└── ...

The folder name format YYYY-MM-DD HH.MM.SS Meeting Name is used to match calendar events.

💡 Usage Examples

Example 1: Weekly Meeting Transcripts

# Transcribe all recordings from last week
python video_transcriber.py ~/Documents/Zoom --model base

# Review transcripts in Google Docs
# Speaker names automatically pulled from calendar

Example 2: Client Call Archive

# Process all client recordings with better accuracy
python video_transcriber.py ~/Videos/ClientCalls \
  --model medium \
  --folder-id abc123xyz

# All transcripts organized in specific Drive folder

Example 3: Conference Recording

# Use AI speaker diarization for multi-speaker panel
export HF_TOKEN=hf_your_token
python video_transcriber.py ~/Conferences/2024 \
  --model medium \
  --recursive

Example 4: Incremental Processing

# Only process videos from this week
python video_transcriber.py ~/Documents/Zoom --since 2024-12-01

# Process videos from a specific date and time
python video_transcriber.py ~/Documents/Zoom --since "2024-12-01 14:30:00"

# Useful for daily/weekly automation - only transcribe new recordings
python video_transcriber.py ~/Documents/Zoom --since $(date -v-7d +%Y-%m-%d)

🔒 Security & Privacy

• ✅ All AI processing is local - Videos never sent to external servers
• ✅ No OpenAI API calls - Zero data sent to cloud
• ✅ Google OAuth - Secure authentication flow
• ✅ Minimal permissions - Only Docs/Drive/Calendar access
• ✅ Token stored locally - credentials.json and token.pickle stay on your machine

🐛 Troubleshooting

"ffmpeg not found"

brew install ffmpeg  # macOS
sudo apt install ffmpeg  # Ubuntu/Debian

"Calendar API has not been used"

Enable Calendar API in Google Cloud Console: https://console.cloud.google.com/apis/library/calendar-json.googleapis.com

"No calendar event found"

Check that:

• Video folder follows Zoom naming: YYYY-MM-DD HH.MM.SS Meeting Name
• Calendar event exists within ±30 minutes of recording time
• Calendar API is enabled and authenticated

Slow processing

• Use smaller model (--model base or --model tiny)
• Enable GPU if available (automatic)
• Process overnight for large batches

Incorrect speaker names

• Verify calendar event has attendees listed
• Try AI diarization with --hf-token for better accuracy
• Check that Zoom folder timestamp matches meeting time

📝 Cost Comparison

Solution	Cost	Processing Speed	Accuracy
This Tool (Zoom2GoogleTranscript)	$0	Local (2-10x realtime)	High
Whisper API	$0.006/min	Very Fast	High
Google Speech-to-Text	$0.016/min	Very Fast	Medium
Rev.ai	$1.50/min	Fast	Very High

100 hours of video:

• Zoom2GoogleTranscript: $0
• Whisper API: $36
• Google Speech-to-Text: $96
• Rev.ai: $9,000

🤝 Contributing

Contributions welcome! Areas for improvement:

• Support for additional video formats (mov, avi, webm)
• Parallel processing for faster batch jobs
• Custom speaker name mapping
• Integration with other calendar systems
• Improved speaker diarization algorithms

📄 License

MIT License - Free for personal and commercial use.

👤 Author

Stephen Sklarew (@stephenhsklarew)

🙏 Acknowledgments

• OpenAI Whisper - Speech recognition model
• pyannote.audio - Speaker diarization
• Google APIs - Docs, Drive, Calendar integration

📞 Support

For issues or questions:

• Open an issue on GitHub
• Email: stephen@synaptiq.ai