Update app.py

app.py

@@ -10,15 +10,44 @@ import gradio as gr
 from PIL import Image
 import pytesseract
 from pdf2image import convert_from_path
+import speech_recognition as sr
 
 app = FastAPI(title="Free MarkItDown Service")
 md = MarkItDown()
 
 IMAGE_EXTENSIONS = {".png", ".jpg", ".jpeg", ".bmp", ".tiff", ".webp"}
+AUDIO_EXTENSIONS = {".mp3", ".wav", ".m4a", ".ogg", ".flac"}
 LEGACY_OFFICE_EXTENSIONS = {".doc", ".xls", ".ppt", ".rtf"}
 
+def process_audio_file(file_path: str) -> str:
+    """Transcribes audio files locally using speech_recognition and ffmpeg."""
+    file_ext = os.path.splitext(file_path)[1].lower()
+    wav_path = file_path
+
+    # Convert non-WAV audio formats to WAV using ffmpeg
+    if file_ext != ".wav":
+        wav_path = f"/tmp/{uuid4()}.wav"
+        subprocess.run(
+            ["ffmpeg", "-y", "-i", file_path, "-ac", "1", "-ar", "16000", wav_path],
+            check=True,
+            stdout=subprocess.DEVNULL,
+            stderr=subprocess.DEVNULL
+        )
+
+    recognizer = sr.Recognizer()
+    try:
+        with sr.AudioFile(wav_path) as source:
+            audio_data = recognizer.record(source)
+            text = recognizer.recognize_sphinx(audio_data)
+            return f"## Audio Transcription\n\n{text}" if text.strip() else "Audio processed, but no speech was detected."
+    except Exception as e:
+        return f"Audio Transcription Error: {str(e)}"
+    finally:
+        if wav_path != file_path and os.path.exists(wav_path):
+            os.remove(wav_path)
+
 def run_free_ocr(file_path: str) -> str:
-    """Fallback OCR pipeline using Tesseract."""
+    """Fallback OCR pipeline using Tesseract (images and PDFs only)."""
     file_ext = os.path.splitext(file_path)[1].lower()
     
     if file_ext in IMAGE_EXTENSIONS:
@@ -55,7 +84,7 @@ def process_input(source_path_or_url: str, original_filename: str = None) -> str
 
     target = source_path_or_url.strip()
 
-    # If HTTP/HTTPS URL
+    # Route URLs
     if target.startswith("http://") or target.startswith("https://"):
         try:
             res = md.convert(target)
@@ -63,10 +92,8 @@ def process_input(source_path_or_url: str, original_filename: str = None) -> str
         except Exception as e:
             return f"URL Conversion Error: {str(e)}"
 
-    # 1. Detect extension from the original filename (e.g., 'document.doc')
     ext = os.path.splitext(original_filename or target)[1].lower()
     
-    # 2. FORCE the temp file on disk to have the real extension so MarkItDown knows what it is!
     working_path = target
     temp_with_ext = None
     if ext and not target.lower().endswith(ext):
@@ -77,40 +104,43 @@ def process_input(source_path_or_url: str, original_filename: str = None) -> str
     temp_converted_office = None
 
     try:
-        # Route Images directly to Tesseract
+        # Route 1: Audio files -> Local Speech Recognition
+        if ext in AUDIO_EXTENSIONS:
+            return process_audio_file(working_path)
+
+        # Route 2: Images -> Tesseract OCR
         if ext in IMAGE_EXTENSIONS:
             return run_free_ocr(working_path)
 
-        # Route Legacy Office files (.doc) to LibreOffice first
+        # Route 3: Legacy Office (.doc, .xls) -> LibreOffice
         if ext in LEGACY_OFFICE_EXTENSIONS:
             temp_converted_office = convert_legacy_office_file(working_path, ext)
             working_path = temp_converted_office
 
-        # Standard MarkItDown extraction
+        # Route 4: Standard MarkItDown extraction (PDFs, DOCX, XLSX, PPTX, HTML, TXT)
         result = md.convert(working_path)
         text_content = result.text_content if result else ""
 
-        # Fallback to Tesseract if extraction yielded blank text
-        if not text_content.strip():
+        if not text_content.strip() and ext in {".pdf"}:
             text_content = run_free_ocr(working_path)
 
         return text_content if text_content.strip() else "Unable to extract text from document."
 
     except Exception as e:
-        print(f"MarkItDown failed on '{working_path}'. Executing Tesseract OCR fallback...")
-        try:
-            ocr_text = run_free_ocr(working_path)
-            if ocr_text and ocr_text.strip():
-                return ocr_text
-        except Exception:
-            pass
+        # OCR fallback ONLY for documents/images, never audio
+        if ext in IMAGE_EXTENSIONS or ext == ".pdf":
+            try:
+                ocr_text = run_free_ocr(working_path)
+                if ocr_text and ocr_text.strip():
+                    return ocr_text
+            except Exception:
+                pass
 
         print("--- CONVERSION ERROR ---")
         traceback.print_exc()
         return f"Conversion Error: File type '{ext}' could not be processed."
 
     finally:
-        # Cleanup temporary files
         if temp_with_ext and os.path.exists(temp_with_ext):
             os.remove(temp_with_ext)
         if temp_converted_office and os.path.exists(temp_converted_office):
@@ -149,7 +179,6 @@ async def convert_file_api(file: UploadFile = None, url: str = None):
 # -------------------------------------------------------------
 def gradio_handler(file_obj, url_input):
     if file_obj is not None:
-        # Extract the real filename (e.g. "my_doc.pdf" instead of "/tmp/gradio/xyz123")
         orig_name = getattr(file_obj, "orig_name", file_obj.name)
         return process_input(file_obj.name, original_filename=orig_name)
     elif url_input and url_input.strip():
@@ -157,11 +186,11 @@ def gradio_handler(file_obj, url_input):
     return "Please upload a document or enter a valid URL."
 
 with gr.Blocks(title="Free MarkItDown Tester") as demo:
-    gr.Markdown("# 📄 Free Document & URL Converter")
+    gr.Markdown("# 📄 Free Document & Audio Converter")
     
     with gr.Row():
         with gr.Column():
-            file_input = gr.File(label="Upload File")
+            file_input = gr.File(label="Upload File (.pdf, .doc, .mp3, .png, etc.)")
             url_input = gr.Textbox(label="Or enter URL", placeholder="https://example.com/document.pdf")
             submit_btn = gr.Button("Convert to Markdown", variant="primary")