Upgrade: Hybrid Digital/OCR scanner for 100% accurate ebook reading

app.py

@@ -8,7 +8,8 @@ from pipeline.tts import generate_tamil_speech
 from pipeline.document_parser import (
     extract_text_from_document, 
     get_pdf_page_as_image, 
-    get_pdf_page_count
+    get_pdf_page_count,
+    get_text_from_page
 )
 
 # Setup logging
@@ -64,12 +65,18 @@ def load_comic_page(pdf_path, page_num):
     status = f"Page {page_num + 1} of {total_pages}"
     return img_path, status, page_num
 
-def process_comic_page(img_path, emotion_choice):
+def process_comic_page(img_path, pdf_path, page_num, emotion_choice):
     if not img_path:
         return "No page loaded", "", None
     
-    # OCR on the rendered comic page
-    extracted_text = extract_text_from_image(img_path)
+    # 1. Try Direct Digital Extraction (100% Accuracy for ebooks/EPUBs)
+    extracted_text = get_text_from_page(pdf_path, page_num)
+    
+    # 2. Fallback to AI OCR (For image-based comics)
+    if not extracted_text or len(extracted_text.strip()) < 5:
+        print(f"DEBUG: No digital text found. Falling back to AI OCR...")
+        extracted_text = extract_text_from_image(img_path)
+        
     if not extracted_text.strip():
         return "No text found on this page", "", None
         
@@ -137,7 +144,7 @@ with gr.Blocks(title="Tamil Comic & Manga Reader AI") as demo:
     
     read_page_btn.click(
         process_comic_page, 
-        inputs=[comic_display, voice_style_comic], 
+        inputs=[comic_display, comic_pdf_path, current_page, voice_style_comic], 
         outputs=[comic_text, comic_tamil, comic_audio]
     )
 
@@ -172,8 +179,8 @@ with gr.Blocks(title="Tamil Comic & Manga Reader AI") as demo:
         if not img: # End of book
             return gr.update(), status, p_num, gr.update(), gr.update(), gr.update()
             
-        # 2. Process the new page
-        txt, tam, aud = process_comic_page(img, voice)
+        # 2. Process the new page (Using Hybrid Mode)
+        txt, tam, aud = process_comic_page(img, pdf, p_num, voice)
         return img, status, p_num, txt, tam, aud
 
     # The hidden button triggers the actual logic

pipeline/document_parser.py

@@ -62,6 +62,20 @@ def get_pdf_page_as_image(file_path: str, page_num: int) -> str:
         print(f"ERROR: PDF rendering failed: {e}")
         return None
 
+def get_text_from_page(file_path: str, page_num: int) -> str:
+    """
+    Tries to extract digital text directly from a specific page.
+    """
+    try:
+        doc = fitz.open(file_path)
+        if page_num >= len(doc):
+            return ""
+        text = doc[page_num].get_text().strip()
+        doc.close()
+        return text
+    except:
+        return ""
+
 def extract_text_from_document(file_path: str) -> str:
     """
     Dispatcher to extract text based on file extension.