app.py

import gradio as gr
from paddleocr import PaddleOCR
import numpy as np
import os
from langdetect import detect
from openai import OpenAI
from transformers import pipeline

# Initialize PaddleOCR
ocr_reader = PaddleOCR(use_angle_cls=True, lang='en')

# Initialize Whisper Model via Hugging Face Transformers
whisper_model = pipeline(
    task="automatic-speech-recognition",
    model="openai/whisper-small",
    device=0
)

# Initialize OpenAI Client
client = OpenAI(api_key=os.getenv("OPENAI_API_KEY"))

def detect_language(text):
    try:
        lang = detect(text)
    except:
        lang = "unknown"
    return lang

def gpt_clean_and_translate(text, target_language):
    if not text.strip():
        return "No text detected.", ""

    prompt = f"""
You are an expert document reader and translator. You will receive noisy extracted text from a government ID. Your tasks:

1. Identify and extract these fields: Name, Address, Date of Birth, Expiry Date, Class, Sex.
2. Output the information in full English sentences.
3. Translate the full text into {target_language}.
If the target language is English, just output clean English sentences.
"""

    response = client.chat.completions.create(
        model="gpt-4o",
        messages=[
            {"role": "system", "content": prompt},
            {"role": "user", "content": text}
        ],
        temperature=0.2
    )

    cleaned_translation = response.choices[0].message.content.strip()
    return cleaned_translation

def process_document(image, target_language, language_group):
    if not isinstance(image, np.ndarray):
        image = np.array(image)

    # OCR - Text Extraction using PaddleOCR
    ocr_result = ocr_reader.ocr(image)

    extracted_texts = []
    for line in ocr_result[0]:
        text = line[1][0]
        extracted_texts.append(text)

    extracted_text = " ".join(extracted_texts)

    # Language Detection
    source_language = detect_language(extracted_text)

    # GPT Cleaning and Translation
    translation = gpt_clean_and_translate(extracted_text, target_language)

    return extracted_text, source_language, translation

def process_audio(audio, target_language):
    # Speech Recognition
    result = whisper_model(audio)
    extracted_text = result['text']

    # Language Detection
    source_language = detect_language(extracted_text)

    # GPT Cleaning and Translation
    translation = gpt_clean_and_translate(extracted_text, target_language)

    return extracted_text, source_language, translation

# Gradio Interface
document_interface = gr.Interface(
    fn=process_document,
    inputs=[
        gr.Image(type="pil", label="Upload a Document Image (e.g., Passport, ID, Government Form)"),
        gr.Radio(choices=["English", "Arabic"], label="Translate To"),
        gr.Dropdown(choices=["Arabic", "Russian", "Other (French, English)"], label="Document Language Group")
    ],
    outputs=[
        gr.Textbox(label="Extracted Text"),
        gr.Textbox(label="Detected Source Language"),
        gr.Textbox(label="Translated and Structured Text")
    ],
    title="🚨 Police Vision & Translator - Document Scanner",
    description="Upload an image document. The system will auto-detect the source language and generate clean translated output."
)

audio_interface = gr.Interface(
    fn=process_audio,
    inputs=[
        gr.Audio(type="filepath", label="Record Audio (max 30 sec)"),
        gr.Radio(choices=["English", "Arabic"], label="Translate To")
    ],
    outputs=[
        gr.Textbox(label="Transcribed Text"),
        gr.Textbox(label="Detected Source Language"),
        gr.Textbox(label="Translated and Structured Text")
    ],
    title="🚨 Police Vision & Translator - Voice Translator",
    description="Record audio. The system will auto-detect the source language and generate clean translated output."
)

# Combine the Interfaces
app = gr.TabbedInterface(
    [document_interface, audio_interface],
    ["Document Translator", "Voice Translator"]
)

if __name__ == "__main__":
    app.launch(share=True)