Update app.py

app.py

@@ -43,12 +43,12 @@ class ConversationalAI:
             device=self.device
         )
         
-        # Load CORRECT audio emotion recognition model
+        # Load WORKING audio emotion recognition model
         self.emotion_model = pipeline(
             "audio-classification",
-            model="speechbrain/emotion-recognition-wav2vec2-IEMOCAP",
+            model="superb/wav2vec2-base-superb-er",
             device=self.device
-        )[1]
+        )
         
         # Conversation history
         self.conversations = {}
@@ -88,13 +88,29 @@ class ConversationalAI:
             return f"Transcription error: {str(e)}"
     
     def recognize_emotion(self, audio_path):
-        """Recognize emotion from audio using proper audio model"""
+        """Recognize emotion from audio using working model"""
         try:
             if audio_path is None:
                 return "neutral"
             
             result = self.emotion_model(audio_path)
-            return result[0]["label"].lower()
+            emotion_label = result[0]["label"].lower()
+            
+            # Map SUPERB emotions to common emotions
+            emotion_mapping = {
+                "ang": "angry",
+                "hap": "happy", 
+                "exc": "excited",
+                "sad": "sad",
+                "fru": "frustrated",
+                "fea": "fearful",
+                "sur": "surprised",
+                "neu": "neutral",
+                "dis": "disgusted"
+            }
+            
+            return emotion_mapping.get(emotion_label, emotion_label)
+            
         except Exception as e:
             print(f"Emotion recognition error: {e}")
             return "neutral"
@@ -105,9 +121,25 @@ class ConversationalAI:
             if text.startswith("Transcription error") or not text.strip():
                 return "I'm sorry, I couldn't understand what you said. Could you please try again?"
             
-            # Build context-aware prompt
-            emotion_prompt = f"[User seems {emotion}] " if emotion != "neutral" else ""
-            prompt = f"{emotion_prompt}User: {text}\nMaya:"
+            # Build context-aware prompt with emotion
+            emotion_responses = {
+                "angry": "I understand you're feeling frustrated. Let me help you with that.",
+                "sad": "I can sense you're feeling down. I'm here to listen and support you.",
+                "happy": "I love your positive energy! That's wonderful to hear.",
+                "excited": "Your enthusiasm is contagious! Tell me more about it.",
+                "fearful": "I can hear the concern in your voice. Let's work through this together.",
+                "surprised": "That sounds quite unexpected! What happened?",
+                "frustrated": "I can tell this is bothering you. Let's see how I can help.",
+                "neutral": "I'm listening. Please go on."
+            }
+            
+            emotion_context = emotion_responses.get(emotion, "I'm here to help.")
+            
+            # Simple but effective response generation
+            if len(text.split()) < 3:
+                return f"{emotion_context} Could you tell me more about that?"
+            
+            prompt = f"User ({emotion}): {text}\nMaya (helpful assistant):"
             
             # Tokenize with proper attention mask
             inputs = self.llm_tokenizer(
@@ -122,7 +154,7 @@ class ConversationalAI:
                 outputs = self.llm_model.generate(
                     input_ids=inputs.input_ids,
                     attention_mask=inputs.attention_mask,
-                    max_new_tokens=80,
+                    max_new_tokens=60,
                     temperature=0.7,
                     do_sample=True,
                     pad_token_id=self.llm_tokenizer.eos_token_id,
@@ -135,9 +167,9 @@ class ConversationalAI:
                 skip_special_tokens=True
             ).strip()
             
-            # Clean up response
-            if not response:
-                response = "I understand. Could you tell me more about that?"
+            # Clean up and add emotion context if response is empty
+            if not response or len(response) < 5:
+                return emotion_context
             
             return response
             
@@ -336,7 +368,7 @@ with gr.Blocks(
         outputs=[transcription_output, audio_output, conversation_history]
     )
 
-# Launch the app - FIXED: Removed show_tips parameter
+# Launch the app
 if __name__ == "__main__":
     demo.launch(
         server_name="0.0.0.0",