updated the app and nodes to help debug the streaming issue to hugging face. full pipeline works locally on docker but slight issue with huggingface.

app.py

@@ -1,11 +1,32 @@
 import os
 import sys
+import streamlit as st
 
 # Add the package directory to Python path
 sys.path.append(os.path.dirname(os.path.abspath(__file__)))
 
+# Configure Streamlit page before importing the main app
+st.set_page_config(
+    page_title="HypeCast Generator",
+    page_icon="🎙️",
+    layout="wide",
+    initial_sidebar_state="collapsed"
+)
+
 # Import and run the Streamlit app
 from hype_pack.streamlit_app import main
 
+# Remove the duplicate page config from streamlit_app.py
 if __name__ == "__main__":
+    # Enable CORS for audio streaming
+    headers = {
+        "Access-Control-Allow-Origin": "*",
+        "Access-Control-Allow-Methods": "GET, POST, OPTIONS",
+        "Access-Control-Allow-Headers": "Content-Type"
+    }
+    
+    # Initialize session state if needed
+    if 'audio_bytes' not in st.session_state:
+        st.session_state.audio_bytes = None
+        
     main() 

hype_pack/utils/nodes.py

@@ -216,26 +216,41 @@ async def text_to_speech_node(interview_state: InterviewState, selected_speaker:
 
         text_content = str(interview_state.transcript.content)
         print(f"Converting text: {text_content[:100]}...")
+        print(f"Using voice_id: {voice_id}")
+        print(f"LMNT API Key present: {'Yes' if os.getenv('LMNT_API_KEY') else 'No'}")
 
         async with speech as s:
-            result = await s.synthesize(
-                text=text_content,
-                voice=voice_id,
-                format='mp3',
-                sample_rate=24000,
-                speed=1.0
-            )
-            print(f"API Response received. Size: {len(result)} bytes")
-
-        # Store bytes directly in interview_state instead of writing to file
+            try:
+                result = await s.synthesize(
+                    text=text_content,
+                    voice=voice_id,
+                    format='mp3',
+                    sample_rate=24000,
+                    speed=1.0
+                )
+                print(f"API Response received. Size: {len(result)} bytes")
+                print(f"Response type: {type(result)}")
+                
+                if not result or len(result) < 100:  # Assuming valid MP3 files are larger than 100 bytes
+                    raise ValueError(f"Invalid audio response size: {len(result)} bytes")
+                
+            except Exception as api_error:
+                print(f"LMNT API Error: {str(api_error)}")
+                raise
+
+        # Store bytes directly in interview_state
         if result and isinstance(result, bytes):
             interview_state.audio_bytes = result
-            print(f"Stored {len(result)} bytes in interview_state")
+            print(f"Successfully stored {len(result)} bytes in interview_state")
+        else:
+            raise ValueError(f"Unexpected response type: {type(result)}")
         
     except Exception as e:
         print(f"Error in text-to-speech conversion: {str(e)}")
         logging.error(f"Error in text-to-speech conversion: {str(e)}")
         interview_state.audio_bytes = None
+        # Optionally, set an error message that can be displayed in the UI
+        st.error(f"Failed to generate audio: {str(e)}")
         
     return interview_state