utils.py

import gdown
import os
import re
from docx import Document
from datetime import datetime

def format_timestamp(seconds):
    if seconds is None: return "00:00"
    td = int(seconds)
    minutes = td // 60
    secs = td % 60
    return f"{minutes:02d}:{secs:02d}"

def download_from_gdrive(url):
    try:
        if not os.path.exists("temp_downloads"): os.makedirs("temp_downloads")
        output_path = f"temp_downloads/drive_audio_{datetime.now().strftime('%Y%m%d_%H%M%S')}.mp3"
        return gdown.download(url, output_path, quiet=False, fuzzy=True)
    except Exception as e:
        print(f"Error: {e}")
        return None

def clean_text(text):
    """
    Uses Regex to find and remove repeating phrases like 'कर दो कर दो कर दो'.
    """
    # Removes repetitions of 3 or more identical words
    text = re.sub(r'(\b\w+\b)( \1){2,}', r'\1', text)
    return text

def save_as_docx(transcripts):
    doc = Document()
    doc.add_heading('Hindi Transcription Report (Cleaned)', 0)
    
    for entry in transcripts:
        doc.add_heading(f"File: {entry['filename']}", level=1)
        content = entry['text']
        
        if isinstance(content, list):
            last_text = ""
            for chunk in content:
                text = clean_text(chunk['text'])
                
                # Skip if it's a hallucination (very high repetition in small chunk)
                if text == last_text or len(text) < 2:
                    continue
                
                ts = chunk.get('timestamp', (0, 0))
                p = doc.add_paragraph()
                p.add_run(f"[{format_timestamp(ts[0])}] ").bold = True
                p.add_run(text)
                last_text = text
        else:
            doc.add_paragraph(clean_text(content))
            
        doc.add_page_break()
    
    report_name = f"Transcripts_Cleaned_{datetime.now().strftime('%Y%m%d_%H%M%S')}.docx"
    doc.save(report_name)
    return report_name