Update app.py

app.py

@@ -1,26 +1,25 @@
 import os
 import json
+import asyncio
+import nest_asyncio
+import edge_tts
 from dotenv import load_dotenv
 from pypdf import PdfReader
 import gradio as gr
-from elevenlabs.client import ElevenLabs
 from huggingface_hub import InferenceClient
 from spaces import GPU 
 
+# Allow async loops in Gradio
+nest_asyncio.apply()
+
 # Load environment keys
 load_dotenv()
-ELEVEN_API_KEY = os.getenv("ELEVEN_API_KEY")
 HF_TOKEN = os.getenv("HF_TOKEN") # Automatically set in Spaces
 
-# Initialize Clients
+# Initialize Client (Qwen 72B)
 if HF_TOKEN:
-    # We use the 72B Instruct model for best reasoning/JSON performance
     hf_client = InferenceClient(model="Qwen/Qwen2.5-72B-Instruct", token=HF_TOKEN)
 
-client = None
-if ELEVEN_API_KEY:
-    client = ElevenLabs(api_key=ELEVEN_API_KEY)
-
 # =========================
 # HELPER FUNCTIONS
 # =========================
@@ -36,6 +35,12 @@ def extract_text_from_pdf(pdf):
     except Exception as e:
         return f"Error reading PDF: {e}"
 
+async def generate_audio_file(text, voice, output_path):
+    """Generates audio using free Edge TTS"""
+    communicate = edge_tts.Communicate(text, voice)
+    await communicate.save(output_path)
+    return output_path
+
 # =========================
 # CORE LOGIC
 # =========================
@@ -52,8 +57,8 @@ def generate_script(pdf_file, persona_style):
     if not pdf_file:
         return "⚠️ Upload a PDF first.", None, new_state
 
-    if not HF_TOKEN or not ELEVEN_API_KEY:
-        return "⚠️ Missing Keys (HF_TOKEN or ELEVEN_API_KEY).", None, new_state
+    if not HF_TOKEN:
+        return "⚠️ Missing HF_TOKEN. This usually works automatically in Spaces.", None, new_state
 
     pdf_text = extract_text_from_pdf(pdf_file)
     new_state["full_text"] = pdf_text
@@ -63,7 +68,7 @@ def generate_script(pdf_file, persona_style):
         "Gossip Columnist": "You are a gossip columnist host. Tone: Dramatic, sensationalist, and excited.",
     }
 
-    # Qwen System Prompt setup
+    # Qwen System Prompt
     system_instruction = f"""
     {prompts.get(persona_style)}
     
@@ -105,8 +110,8 @@ def generate_script(pdf_file, persona_style):
     except Exception as e:
         return f"Error with Qwen: {e}", None, new_state
 
-@GPU
-def play_next_chunk(state_data):
+# We use async here for Edge TTS
+async def play_next_chunk(state_data):
     if not state_data or not state_data.get("script"):
         return None, "⚠️ No script generated yet.", state_data
 
@@ -118,21 +123,20 @@ def play_next_chunk(state_data):
 
     line = script[idx]
 
-    voice_id = "nPczCjz82tPNOwVbpGE2" # Default voice
-    if state_data["persona"] == "Gossip Columnist" and line["speaker"] == "Host B":
-        voice_id = "21m00Tcm4TlvDq8ikWAM"
+    # SELECT VOICES (Free Edge TTS)
+    # Host A = Male, Host B = Female
+    voice_id = "en-US-ChristopherNeural" 
+    
+    if line["speaker"] == "Host B":
+        voice_id = "en-US-AriaNeural"
+    
+    # Switch voices for Gossip mode
+    if state_data["persona"] == "Gossip Columnist":
+        voice_id = "en-US-EricNeural" if line["speaker"] == "Host A" else "en-US-AnaNeural"
 
     try:
-        audio_stream = client.generate(
-            text=line["text"],
-            voice=voice_id,
-            model="eleven_monolingual_v1"
-        )
-
         save_path = f"temp_{idx}.mp3"
-        with open(save_path, "wb") as f:
-            for chunk in audio_stream:
-                f.write(chunk)
+        await generate_audio_file(line["text"], voice_id, save_path)
 
         state_data["current_index"] += 1
         return save_path, f"{line['speaker']}: {line['text']}", state_data
@@ -140,8 +144,7 @@ def play_next_chunk(state_data):
     except Exception as e:
         return None, f"Audio error: {e}", state_data
 
-@GPU
-def interrupt_and_ask(question, state_data):
+async def interrupt_and_ask(question, state_data):
     if not state_data or not state_data.get("full_text"):
         return None, "Upload PDF first.", state_data
 
@@ -159,16 +162,8 @@ def interrupt_and_ask(question, state_data):
         return None, f"Qwen Error: {e}", state_data
 
     try:
-        audio_stream = client.generate(
-            text=answer,
-            voice="nPczCjz82tPNOwVbpGE2",
-            model="eleven_monolingual_v1"
-        )
-
         save_path = "interrupt.mp3"
-        with open(save_path, "wb") as f:
-            for chunk in audio_stream:
-                f.write(chunk)
+        await generate_audio_file(answer, "en-US-ChristopherNeural", save_path)
 
         return save_path, answer, state_data
     except Exception as e: