Upload 2 files

src/models/audio_classifier.py

@@ -214,16 +214,22 @@ class AudioEnsemblePipeline:
         
         prob_dict = {str(self.classes_[i]): round(float(probs[i]), 4) for i in range(len(self.classes_))}
         
-        high_stress_emotions = ["Stress", "Anxiety", "Depression", "Emotional Distress"]
+        high_stress_emotions = ["Angry", "Fearful", "Sad", "Disgust"]
         
-        # Only sum probabilities for categories that actually exist in self.classes_
+        # Safe prob sum calculation checking if classes exist
         stress_prob_sum = 0.0
         for e in high_stress_emotions:
             if e in self.classes_:
                 stress_prob_sum += probs[self.classes_.index(e)]
-        
+                
         rms_val = float(feature_vector_195[-1])
-        stress_intensity = round(float(min(100.0, max(5.0, stress_prob_sum * 85.0 + (rms_val * 60.0)))), 2)
+        
+        # Reduce the impact of RMS volume so normal speech doesn't get flagged as Stress
+        # Default stress relies much more on the predicted probabilities
+        base_intensity = stress_prob_sum * 100.0
+        volume_penalty = min(20.0, rms_val * 20.0) # Cap volume contribution
+        
+        stress_intensity = round(float(min(100.0, max(5.0, base_intensity + volume_penalty))), 2)
         
         return {
             "predicted_emotion": pred_emotion,

src/models/fusion_engine.py

@@ -64,17 +64,17 @@ class LateDecisionFusion:
         
         # Determine primary stress origin
         t_cat = text_result.get("predicted_category", "Normal")
-        if combined_stress_score < 30.0:
+        if combined_stress_score < 40.0:
             final_category = "Normal"
         else:
             final_category = t_cat if t_cat != "Normal" else "Stress"
             
         # Determine clinical risk tier
-        if combined_stress_score < 30.0:
+        if combined_stress_score < 40.0:
             risk_tier = "Minimal / Normal"
             color_code = "green"
             action_summary = "No significant psychological stress detected. Emotional tone is balanced."
-        elif combined_stress_score < 55.0:
+        elif combined_stress_score < 60.0:
             risk_tier = f"Mild {final_category}" if final_category not in ["Normal", "Stress", "Calm / Normal"] else "Mild Stress"
             color_code = "blue"
             action_summary = f"Mild symptoms of {final_category.lower()} observed. Recommended: short breaks and time-management strategies."
@@ -132,10 +132,18 @@ class LateDecisionFusion:
     def _single_modality_audio_result(self, audio_result):
         score = audio_result.get("acoustic_stress_score", 0.0)
         emotion = audio_result.get("predicted_emotion", "Neutral")
-        cat = "Non-Academic / Vocal Stress" if emotion in ["Angry", "Fearful", "Sad", "Disgust"] else "Calm / Normal"
-        if score < 30.0:
+        
+        # Determine category based on emotion or score
+        if emotion in ["Angry", "Fearful", "Sad", "Disgust", "Stress", "Anxiety", "Depression", "Emotional Distress"]:
+            cat = "Non-Academic / Vocal Stress"
+        elif score >= 40.0:
+            cat = "Non-Academic / Vocal Stress"
+        else:
+            cat = "Calm / Normal"
+            
+        if score < 40.0:
             tier, color, summary = "Minimal / Normal", "green", "Vocal tone is calm and stable."
-        elif score < 55.0:
+        elif score < 60.0:
             tier, color, summary = "Mild Stress", "blue", f"Slight vocal tension ({emotion.lower()}) observed."
         elif score < 80.0:
             tier, color, summary = "Moderate Stress", "orange", f"High vocal agitation ({emotion.lower()}) and spectral energy detected."