Update text_ai.py

text_ai.py

@@ -1,69 +1,32 @@
-# text_ai.py
-import speech_recognition as sr
-from deep_translator import GoogleTranslator
-from transformers import pipeline, AutoTokenizer, AutoModelForSequenceClassification
-import os
+from transformers import pipeline
+import config as cfg
 
 class TextEngine:
     def __init__(self):
-        print("[Text AI] Initializing Cloud DeBERTa Model...")
+        # Using a specialized emotion model
+        self.classifier = pipeline("text-classification", 
+                                   model="j-hartmann/emotion-english-distilroberta-base", 
+                                   return_all_scores=True)
         
-        # Don't forget to initialize the speech recognizer!
-        self.recognizer = sr.Recognizer()
-        
-        model_id = "cross-encoder/nli-deberta-v3-small"
+        # Map model labels to TraceMind internal labels
+        self.mapping = {
+            'anger': 'Angry',
+            'disgust': 'Disgust',
+            'fear': 'Fear',
+            'joy': 'Happy',
+            'neutral': 'Neutral',
+            'sadness': 'Sad',
+            'surprise': 'Surprise'
+        }
+    
+    def analyze_text(self, text):
+        if not text:
+            return {label: 0.0 for label in ['Angry', 'Disgust', 'Fear', 'Happy', 'Neutral', 'Sad', 'Surprise']}
         
-        try:
-            self.tokenizer = AutoTokenizer.from_pretrained(model_id, use_fast=False)
-            self.model = AutoModelForSequenceClassification.from_pretrained(model_id)
-            
-            print("[Text AI] Engine Ready & Loaded Successfully.")
-        except Exception as e:
-            print(f"❌ CRITICAL ERROR: Failed to fetch DeBERTa: {e}")
+        raw_results = self.classifier(text)[0]
+        # Convert and capitalize labels to match ai.py
+        return {self.mapping[item['label']]: item['score'] for item in raw_results}
 
     def transcribe_and_translate(self, audio_path):
-        """ Reads audio file -> Text -> English Text """
-        try:
-            with sr.AudioFile(audio_path) as source:
-                audio_data = self.recognizer.record(source)
-            
-            # Speech to Text
-            text = self.recognizer.recognize_google(audio_data)
-            print(f"[Text AI] Transcribed: {text}")
-            
-            # Translate to English (if needed)
-            translated_text = self.translator.translate(text)
-            if translated_text != text:
-                print(f"[Text AI] Translated: {translated_text}")
-            
-            return translated_text
-            
-        except sr.UnknownValueError:
-            print("[Text AI] Audio was silent or unintelligible.")
-            return ""
-        except Exception as e:
-            print(f"[Text AI] Audio Error: {e}")
-            return ""
-
-    def analyze_text(self, text):
-        """ Analyzes the emotion of the English text """
-        if not text or self.classifier is None:
-            return None
-
-        # 1. Run the Model
-        results = self.classifier(text)[0] 
-        raw_probs = {item['label']: item['score'] for item in results}
-        
-        # 2. Map Professional Labels (lowercase) to Tracemind Labels (Capitalized)
-        # The model uses 'joy', 'anger' -> We need 'Happy', 'Angry'
-        final_probs = {
-            'Angry': raw_probs.get('anger', 0.0),
-            'Disgust': raw_probs.get('disgust', 0.0),
-            'Fear': raw_probs.get('fear', 0.0),
-            'Happy': raw_probs.get('joy', 0.0),    # Map 'joy' to 'Happy'
-            'Neutral': raw_probs.get('neutral', 0.0),
-            'Sad': raw_probs.get('sadness', 0.0),  # Map 'sadness' to 'Sad'
-            'Surprise': raw_probs.get('surprise', 0.0)
-        }
-        
-        return final_probs
+        # (Keep your existing Whisper/Translation logic here)
+        return "Sample transcribed text for demo"