Update app.py

app.py

@@ -1,54 +1,60 @@
 import gradio as gr
-from transformers import pipeline
 from indic_transliteration import sanscript
 from indic_transliteration.sanscript import transliterate
+from transformers import pipeline
 
-#  Emotion Model
+# 1. Load Hindi Emotion Detection Model
 emotion_model = pipeline(
     "text-classification",
-    model="vashuag/HindiEmotion",
-    return_all_scores=True
-)
-
-#  Sentiment Model
-sentiment_model = pipeline(
-    "text-classification",
-    model="LondonStory/txlm-roberta-hindi-sentiment",
+    model="s-nlp/HiEmotions",   # Replace with a public, reliable Hindi emotion model
     return_all_scores=True
 )
 
-# Transliterate Hinglish → Hindi
-def hinglish_to_hindi(text):
+# 2. Basic Hinglish -> Hindi transliteration
+def hinglish_to_hindi(text: str) -> str:
     return transliterate(text, sanscript.ITRANS, sanscript.DEVANAGARI)
 
-# Full Pipeline
-def analyze_text(hinglish_text):
-    raw_hindi = hinglish_to_hindi(hinglish_text)
-    clean_hindi = raw_hindi  # Can insert normalization if needed
+# 3. Normalization layer (common spelling fixes)
+def normalize_hindi(text: str) -> str:
+    corrections = {
+        "उदस्": "उदास",
+        "प्यर्": "प्यार",
+        "भाइयोन्": "भाइयों",
+        "कित्न": "कितना",
+        "टुझे": "तुझे"
+    }
+    for wrong, right in corrections.items():
+        text = text.replace(wrong, right)
+    return text
+
+# 4. Full pipeline
+def analyze_emotion(hinglish_text: str):
+    # Step 1: Transliterate Hinglish → Hindi
+    hindi_text = hinglish_to_hindi(hinglish_text)
+
+    # Step 2: Normalize Hindi text
+    clean_text = normalize_hindi(hindi_text)
 
-    emotions = emotion_model(clean_hindi)[0]
-    sentiments = sentiment_model(clean_hindi)[0]
+    # Step 3: Predict emotion
+    emotion_results = emotion_model(clean_text)
 
+    # Step 4: Find highest emotion
+    top_emotion = max(emotion_results[0], key=lambda x: x['score'])
+
+    # Step 5: Summarize output for user
     return {
         "Hinglish Input": hinglish_text,
-        "Raw Hindi": raw_hindi,
-        "Emotion Scores": {e["label"]: round(e["score"], 3) for e in emotions},
-        "Sentiment Scores": {s["label"]: round(s["score"], 3) for s in sentiments}
+        "Clean Hindi": clean_text,
+        "Predicted Emotion": top_emotion['label'],
+        "Confidence": round(top_emotion['score'], 3)
     }
 
-iface = gr.Interface(
-    fn=analyze_text,
-    inputs=gr.Textbox(label="Enter Hinglish Text"),
-    outputs="json",
-    title="Hinglish → Hindi Emotion & Sentiment Analyzer",
-    description=(
-        "This app uses **documented and public models**:\n"
-        "- **vashuag/HindiEmotion** for emotion classification\n"
-        "- **LondonStory/txlm-roberta-hindi-sentiment** for sentiment (3-class)\n"
-        "Both are transparent, reliable, and well-evaluated on Hindi data."
-    )
-)
-
-if __name__ == "__main__":
-    iface.launch()
+# 5. Gradio UI
+with gr.Blocks() as demo:
+    gr.Markdown("## Hybrid Emotion Detection App (Hinglish → Hindi)")
+    inp = gr.Textbox(label="Enter Hinglish Text")
+    out = gr.JSON(label="Emotion Analysis Result")
+    btn = gr.Button("Analyze Emotion")
+    btn.click(fn=analyze_emotion, inputs=inp, outputs=out)
 
+demo.launch()