README.md

metadata

title: Video Transcription & Analysis
emoji: 🎤
colorFrom: blue
colorTo: purple
sdk: streamlit
sdk_version: 1.28.0
app_file: src/streamlit_app.py
pinned: false
license: mit

🎤 Video Transcription & Analysis

A powerful Streamlit application that transcribes speech from video files and performs comprehensive sentiment and emotion analysis using state-of-the-art pretrained models.

✨ Features

• 🎥 Video Upload - Support for multiple video formats (MP4, AVI, MOV, MKV, WebM, FLV, WMV, M4V)
• 🎤 Speech Transcription - Automatic speech-to-text using OpenAI Whisper
• 👥 Speaker Diarization - Automatically identifies and separates different speakers (Speaker 1, Speaker 2, etc.)
• 😊 Sentiment Analysis - Detects positive, negative, or neutral sentiment for each segment
• 🎭 Emotion Analysis - Identifies emotions: angry, sad, happy, neutral, surprise
• 📊 Visual Analytics - Interactive graphs showing sentiment and emotion distribution
• 📝 Detailed Reports - Per-segment analysis with timestamps and speaker labels
• 📥 Export Options - Download transcription and full analysis reports
• 🌍 Language Detection - Automatic language identification
• ⚡ Multiple Model Sizes - Choose from tiny, base, small, medium, or large Whisper models

🚀 How to Use

1. Upload a Video: Click "Browse files" and select your video file
2. Enable Speaker Diarization: Toggle the checkbox to identify different speakers (enabled by default)
3. Transcribe: Click "🎯 Transcribe Video" button
4. View Results:
   • See speaker-separated transcription
   • View sentiment and emotion analysis
   • Explore interactive graphs
   • Review detailed segment analysis

📊 Analysis Features

Sentiment Analysis

• Positive 😊 - Happy, optimistic, or positive content
• Negative 😞 - Sad, pessimistic, or negative content
• Neutral 😐 - Balanced or neutral content

Emotion Analysis

• Happy 😄 - Joyful and positive emotions
• Sad 😢 - Sadness and melancholy
• Angry 😠 - Anger and frustration
• Neutral 😐 - Balanced emotional state
• Surprise 😲 - Surprise and astonishment

Speaker Diarization

• Automatically detects 2-5 speakers
• Labels them sequentially as Speaker 1, Speaker 2, etc.
• Separates transcription by speaker for easy reading

🛠️ Technical Details

Models Used

Transcription

• OpenAI Whisper - State-of-the-art speech recognition
  • Model sizes: tiny (39M), base (74M), small (244M), medium (769M), large (1550M)
  • Recommended: base (good balance of speed and accuracy)

Sentiment Analysis

• CardiffNLP Twitter-RoBERTa - Pretrained sentiment analysis model
  • Detects: Positive, Negative, Neutral

Emotion Analysis

• j-hartmann Emotion Model - Pretrained emotion detection model
  • Detects: Joy, Sadness, Anger, Fear, Surprise, Disgust, Neutral

Speaker Diarization

• Audio Feature Clustering - Uses MFCC features and hierarchical clustering
  • No external API required
  • Works entirely offline

📋 Requirements

• Python 3.8+
• FFmpeg (automatically installed in Hugging Face Spaces)
• All Python dependencies are listed in requirements.txt

🎯 Use Cases

• Content Analysis - Analyze video content for sentiment and emotions
• Meeting Transcripts - Transcribe and analyze meetings with speaker identification
• Video Editing - Get transcripts for video editing workflows
• Accessibility - Create transcripts for accessibility purposes
• Research - Analyze emotional patterns in video content
• Content Moderation - Detect sentiment and emotions in user-generated content

📝 Notes

• First run will download the models (one-time download, cached for future use)
• Larger videos will take longer to process
• The app automatically extracts audio from video files
• All models are free and pretrained (no API keys required)
• Processing happens entirely in your browser session

🔧 Model Selection

Choose the Whisper model size based on your needs:

• tiny - Fastest, good for quick transcriptions
• base - Recommended balance of speed and accuracy
• small - Better accuracy for important content
• medium - High accuracy for professional use
• large - Best accuracy, slower processing

📄 License

MIT License - Feel free to use and modify for your projects!

🙏 Acknowledgments

• OpenAI for the Whisper model
• Hugging Face for hosting and model distribution
• CardiffNLP for sentiment analysis model
• j-hartmann for emotion detection model

---

Made with ❤️ using Streamlit and Hugging Face Spaces