README.md

metadata

title: Rephrasia - AI-Powered Text Processing API
emoji: 🚀
colorFrom: blue
colorTo: indigo
sdk: docker
app_port: 7860

Rephrasia - AI-Powered Text Processing API

A Flask-based API for text paraphrasing, translation, chat, and more.

Features

• Paraphrasing: Generate multiple paraphrased versions using PEGASUS
• Translation: Bidirectional English ↔ Urdu translation using NLLB-200
• Chat: Conversational AI with DialoGPT maintaining session history
• OCR (Image to Text): Extract text from images and process it
• Batch Processing: Process multiple texts in one request
• Text-to-Speech: Convert text to audio (Urdu/English)
• Export: Save results as PDF or DOCX

System Requirements

• Python 3.9+
• 8GB+ RAM recommended (for models)
• 5GB+ disk space for model weights
• Tesseract OCR installed (for image-to-text)
  • Windows: Download from UB-Mannheim
  • Install to C:\Program Files\Tesseract-OCR\
  • Or adjust path in ocr.py

Installation

# Clone the repository
cd Rephrasia

# Install dependencies
pip install -r requirements.txt

Quick Start

# Run the Flask server
python app.py

# Server starts at http://127.0.0.1:5000

API Endpoints

1. Paraphrasing

Single Text:

POST /api/rephrase
Content-Type: application/json

{
  "text": "Education shapes society."
}

Batch Processing:

POST /api/batch/rephrase
Content-Type: application/json

{
  "texts": ["Text 1", "Text 2", "Text 3"]
}

2. Translation

Single Translation:

POST /api/translate
Content-Type: application/json

{
  "text": "Hello world",
  "language": "urdu"  # or "english"
}

Batch Translation:

POST /api/batch/translate
Content-Type: application/json

{
  "texts": ["Hello", "Goodbye"],
  "language": "urdu"
}

3. Chat

POST /chat
Content-Type: application/json

{
  "text": "Hello, how are you?",
  "session_id": "optional-session-id"
}

4. Text-to-Speech

POST /api/tts
Content-Type: application/json

{
  "text": "یہ ایک ٹیسٹ ہے",
  "language": "urdu"  # or "english"
}

5. Export

Export to PDF:

POST /api/export/pdf
Content-Type: application/json

{
  "original_text": "Original sentence",
  "results": ["Paraphrase 1", "Paraphrase 2"],
  "result_type": "Paraphrased"
}

Export to DOCX:

POST /api/export/docx
Content-Type: application/json

{
  "original_text": "Original sentence",
  "results": ["Translation"],
  "result_type": "Translated"
}

6. OCR (Image to Text)

Extract Text from Image:

POST /api/ocr
Content-Type: multipart/form-data

image: [image file]
language: eng  # or 'urd', 'eng+urd'
preprocess: true  # optional, enhances image quality

OCR + Rephrase (Combined):

POST /api/ocr-rephrase
Content-Type: multipart/form-data

image: [image file]
language: eng
preprocess: true

OCR + Translate (Combined):

POST /api/ocr-translate
Content-Type: multipart/form-data

image: [image file]
ocr_language: eng
target_language: urdu
preprocess: true

Get Supported OCR Languages:

GET /api/ocr/languages

Testing

# Run all tests
pytest

# Run with verbose output
pytest -v

# Run specific test file
pytest tests/test_api.py

PowerShell Examples

# Rephrase
Invoke-RestMethod -Method Post -Uri http://127.0.0.1:5000/api/rephrase `
  -Body (@{text="Education shapes society."} | ConvertTo-Json) `
  -ContentType "application/json"

# Translate
Invoke-RestMethod -Method Post -Uri http://127.0.0.1:5000/api/translate `
  -Body (@{text="Hello"; language="urdu"} | ConvertTo-Json) `
  -ContentType "application/json"

# Batch rephrase
Invoke-RestMethod -Method Post -Uri http://127.0.0.1:5000/api/batch/rephrase `
  -Body (@{texts=@("Text 1", "Text 2")} | ConvertTo-Json) `
  -ContentType "application/json"

# OCR from image
$imagePath = "C:\path\to\image.png"
$form = @{
    image = Get-Item -Path $imagePath
    language = "eng"
    preprocess = "true"
}
Invoke-RestMethod -Method Post -Uri http://127.0.0.1:5000/api/ocr -Form $form

# OCR + Rephrase
Invoke-RestMethod -Method Post -Uri http://127.0.0.1:5000/api/ocr-rephrase -Form $form

Model Information

Feature	Model	Size
Paraphrasing	tuner007/pegasus_paraphrase	~2GB
Translation	facebook/nllb-200-distilled-600M	~2.5GB
Chat	microsoft/DialoGPT-medium	~800MB
TTS	Google Text-to-Speech (gTTS)	API-based
OCR	Tesseract OCR	~100MB (separate install)

Project Structure

Rephrasia/
├── app.py                 # Main Flask application
├── model.py               # Paraphrasing logic
├── translation.py         # Translation logic
├── chat.py                # Chat session management
├── batch_processor.py     # Bulk processing
├── tts.py                 # Text-to-speech
├── export_utils.py        # PDF/DOCX export
├── ocr.py                 # Image to text (OCR)
├── requirements.txt       # Dependencies
├── tests/                 # Test suite
│   ├── test_api.py
│   └── test_chat_manager.py
└── static/                # Generated files
    ├── audio/             # TTS audio files
    └── exports/           # PDF/DOCX exports

Performance Tips

• Models are lazy-loaded on first request
• Use batch endpoints for multiple texts
• Audio/export files auto-cleanup after 24 hours
• Chat sessions are stored in-memory (consider Redis for production)

Troubleshooting

Tesseract Not Found:

• Download Tesseract from here
• Install to default location or update path in ocr.py
• Restart terminal after installation

OCR Not Detecting Text:

• Try with preprocess: true parameter
• Ensure image has clear, readable text
• Check image format (PNG, JPG supported)
• Use correct language code (eng, urd, etc.)

Out of Memory:

• Reduce num_beams in model generation
• Use smaller model variants (DialoGPT-small)
• Process fewer items in batch requests

Slow First Request:

• Models download and load on first use (~5-10 mins)
• Subsequent requests are fast

Import Errors:

• Ensure all requirements are installed: pip install -r requirements.txt
• Python 3.9+ required

License

Open source - free to use and modify

Support

For issues or questions, check the code comments or test files for usage examples.