Info.md

# English Accent Detection Tool

A practical AI tool that analyzes English accents from video content. Built for REM Waste's hiring automation system.

## 🚀 Live Demo

**Deployed App:** [https://accent-detector.streamlit.app](https://accent-detector.streamlit.app)

## Features

- **Video Processing**: Accepts public video URLs (MP4, Loom, etc.)
- **Audio Extraction**: Automatically extracts audio from video files
- **Speech Transcription**: Converts speech to text using Google Speech Recognition
- **Accent Analysis**: Detects English accents with confidence scoring
- **Web Interface**: Simple Streamlit UI for easy testing

## Supported Accents

- American English
- British English  
- Australian English
- Canadian English
- South African English

## Quick Start

### Method 1: Use the Deployed App (Recommended)

1. Visit: [https://accent-detector.streamlit.app](https://accent-detector.streamlit.app)
2. Paste a public video URL
3. Click "Analyze Accent"
4. View results with confidence scores

### Method 2: Local Installation

```bash
# Clone or download the script
git clone <repository-url>
cd accent-detector

# Install dependencies
pip install -r requirements.txt

# Install ffmpeg (required for video processing)
# On macOS:
brew install ffmpeg

# On Ubuntu/Debian:
sudo apt update && sudo apt install ffmpeg

# On Windows:
# Download from https://ffmpeg.org/download.html

# Run the app
streamlit run accent_detector.py
```

## Installation

1. Clone this repository and navigate to the project folder.
2. (Recommended) Create and activate a Python virtual environment:
   ```sh
   python3 -m venv ad_venv
   source ad_venv/bin/activate
   ```
3. Install all dependencies:
   ```sh
   pip install -r requirements.txt
   ```
4. (Optional, but recommended for better performance) Install Watchdog:
   ```sh
   xcode-select --install  # macOS only, for build tools
   pip install watchdog
   ```

## Usage Examples

### Test URLs
```
# Direct MP4 link
https://sample-videos.com/zip/10/mp4/SampleVideo_1280x720_1mb.mp4

# Loom video (public)
https://www.loom.com/share/your-video-id

# Google Drive (public)
https://drive.google.com/file/d/your-file-id/view
```

### Expected Output
```json
{
  "accent": "American",
  "confidence": 78.5,
  "explanation": "High confidence in American accent with strong linguistic indicators.",
  "all_scores": {
    "American": 78.5,
    "British": 23.1,
    "Australian": 15.7,
    "Canadian": 19.2,
    "South African": 8.3
  }
}
```

## Technical Architecture

### Core Components

1. **Video Downloader**: Downloads videos from public URLs
2. **Audio Extractor**: Uses ffmpeg to extract WAV audio
3. **Speech Recognizer**: Google Speech Recognition API
4. **Accent Analyzer**: Pattern matching for linguistic markers
5. **Web Interface**: Streamlit-based UI

### Accent Detection Algorithm

The system analyzes multiple linguistic features:

- **Vocabulary Patterns**: Accent-specific word choices
- **Phonetic Markers**: Pronunciation characteristics  
- **Spelling Patterns**: Regional spelling differences
- **Linguistic Markers**: Characteristic phrases and expressions

### Confidence Scoring

- **0-20%**: Insufficient markers detected
- **21-50%**: Moderate confidence with limited indicators
- **51-75%**: Good confidence with multiple patterns
- **76-100%**: High confidence with strong linguistic evidence

## API Integration

For programmatic access, use the core `AccentDetector` class:

```python
from accent_detector import AccentDetector

detector = AccentDetector()
result = detector.process_video("https://your-video-url.com/video.mp4")

print(f"Accent: {result['accent']}")
print(f"Confidence: {result['confidence']}%")
```

## Deployment

### Streamlit Cloud (Recommended)

1. Fork this repository
2. Connect to Streamlit Cloud
3. Deploy from your GitHub repo
4. Share the public URL

### Docker Deployment

```dockerfile
FROM python:3.9-slim

# Install system dependencies
RUN apt-get update && apt-get install -y ffmpeg

WORKDIR /app
COPY requirements.txt .
RUN pip install -r requirements.txt

COPY . .
EXPOSE 8501

CMD ["streamlit", "run", "accent_detector.py", "--server.port=8501", "--server.address=0.0.0.0"]
```

## Limitations & Considerations

### Current Limitations
- Requires clear speech audio (background noise affects accuracy)
- Works best with 30+ seconds of speech
- Free Google Speech Recognition has daily limits
- Accent detection based on vocabulary/patterns, not phonetic analysis

### Potential Improvements
- Integrate phonetic analysis libraries
- Add more accent varieties (Indian, Irish, etc.)
- Implement batch processing for multiple videos
- Add voice activity detection for better audio segmentation

## Testing

### Manual Testing
1. Test with different accent samples
2. Verify confidence scores are reasonable
3. Check error handling with invalid URLs
4. Test with various video formats

### Automated Testing
```python
def test_accent_detection():
    detector = AccentDetector()
    
    # Test American accent
    american_text = "I'm gonna grab some cookies from the elevator"
    scores = detector.analyze_accent_patterns(american_text)
    assert scores['American'] > scores['British']
    
    # Test British accent  
    british_text = "That's brilliant, quite lovely indeed"
    scores = detector.analyze_accent_patterns(british_text)
    assert scores['British'] > scores['American']
```

## Performance Metrics

- **Video Download**: ~10-30 seconds (depends on file size)
- **Audio Extraction**: ~5-15 seconds
- **Speech Recognition**: ~10-30 seconds
- **Accent Analysis**: <1 second
- **Total Processing**: ~30-90 seconds per video

## Troubleshooting

### Common Issues

**Error: "Could not understand the audio"**
- Solution: Ensure clear speech, minimal background noise

**Error: "Failed to download video"**  
- Solution: Verify URL is public and accessible

**Error: "ffmpeg not found"**
- Solution: Install ffmpeg system dependency

**Low confidence scores**
- Solution: Ensure longer speech samples (30+ seconds)

### Support

For technical issues or feature requests:
1. Check the error messages in the Streamlit interface
2. Verify all dependencies are installed correctly  
3. Test with known working video URLs

## License

MIT License - Free for commercial and personal use.

---

**Built for REM Waste Interview Challenge**  
*Practical AI tools for automated hiring decisions*