Update app.py

app.py

@@ -4,28 +4,35 @@ import pandas as pd
 import plotly.express as px
 
 # ------------------------------
-# Load pretrained models (CPU-friendly, public)
+# Load pretrained models (CPU-friendly)
 # ------------------------------
 text_classifier = pipeline(
     "text-classification",
-    model="j-hartmann/emotion-english-distilroberta-base",  # public
-    top_k=None,   # replaces deprecated return_all_scores
-    device=-1
+    model="j-hartmann/emotion-english-distilroberta-base",
+    top_k=None  # returns all scores
 )
 
+# Use a small, public audio model
 audio_classifier = pipeline(
     "audio-classification",
-    model="superb/wav2vec2-base-superb-er",  # public model
-    device=-1
+    model="superb/wav2vec2-small-superb-er"  # small model
 )
 
 # ------------------------------
 # Map emotion to emoji
 # ------------------------------
 EMOJI_MAP = {
-    "anger": "😡", "disgust": "🤢", "fear": "😨", "joy": "😄",
-    "neutral": "😐", "sadness": "😢", "surprise": "😲",
-    "hap": "😄", "neu": "😐", "sad": "😢", "ang": "😡"
+    "anger": "😡",
+    "disgust": "🤢",
+    "fear": "😨",
+    "joy": "😄",
+    "neutral": "😐",
+    "sadness": "😢",
+    "surprise": "😲",
+    "hap": "😄",  # audio model labels
+    "neu": "😐",
+    "sad": "😢",
+    "ang": "😡"
 }
 
 # ------------------------------
@@ -40,8 +47,8 @@ def fuse_predictions(text_preds=None, audio_preds=None, w_text=0.5, w_audio=0.5)
     scores = {l: 0.0 for l in labels}
 
     def normalize(preds):
-        s = sum(p['score'] for p in preds)
-        return {p['label']: p['score']/s for p in preds}
+        total = sum(p['score'] for p in preds)
+        return {p['label']: p['score']/total for p in preds}
 
     if text_preds:
         t_norm = normalize(text_preds)
@@ -56,7 +63,7 @@ def fuse_predictions(text_preds=None, audio_preds=None, w_text=0.5, w_audio=0.5)
     return {"fused_label": best[0], "fused_score": round(best[1], 3), "all_scores": scores}
 
 # ------------------------------
-# Create bar chart
+# Bar chart function
 # ------------------------------
 def make_bar_chart(scores_dict, title="Emotion Scores"):
     df = pd.DataFrame({
@@ -76,9 +83,10 @@ def make_bar_chart(scores_dict, title="Emotion Scores"):
 def predict(text, audio, w_text, w_audio):
     text_preds, audio_preds = None, None
     if text:
-        text_preds = text_classifier(text)
+        text_preds = text_classifier(text)  # list of dicts
     if audio:
-        audio_preds = audio_classifier(audio)
+        audio_preds = audio_classifier(audio)  # list of dicts
+
     fused = fuse_predictions(text_preds, audio_preds, w_text, w_audio)
 
     # Final emotion with emoji
@@ -86,27 +94,26 @@ def predict(text, audio, w_text, w_audio):
     emoji = EMOJI_MAP.get(label, "")
     final_emotion = f"### Final Predicted Emotion: {label.upper()} {emoji} (score: {fused['fused_score']})"
 
-    # Only show fused chart to avoid Gradio multiple output issues
+    # Fused bar chart
     chart = make_bar_chart(fused['all_scores'], "Fused Emotion Scores")
-
     return final_emotion, chart
 
 # ------------------------------
-# Gradio Interface
+# Build Gradio interface
 # ------------------------------
 with gr.Blocks() as demo:
-    gr.Markdown("## 🎭 Public CPU-Friendly Multimodal Emotion Classifier")
+    gr.Markdown("## 🎭 Multimodal Emotion Classification (Text + Speech)")
 
     with gr.Row():
         with gr.Column():
             txt = gr.Textbox(label="Text input", placeholder="Type something emotional...")
             aud = gr.Audio(type="filepath", label="Upload speech (wav/mp3)")
-            w1 = gr.Slider(minimum=0.0, maximum=1.0, value=0.5, label="Text weight")
-            w2 = gr.Slider(minimum=0.0, maximum=1.0, value=0.5, label="Audio weight")
+            w1 = gr.Slider(minimum=0.0, maximum=1.0, value=0.5, label="Text weight (w_text)")
+            w2 = gr.Slider(minimum=0.0, maximum=1.0, value=0.5, label="Audio weight (w_audio)")
             btn = gr.Button("Predict")
         with gr.Column():
-            final_label = gr.Markdown()
-            chart_output = gr.Plot()
+            final_label = gr.Markdown(label="Predicted Emotion")
+            chart_output = gr.Plot(label="Fused Emotion Scores")
 
     btn.click(fn=predict, inputs=[txt, aud, w1, w2], outputs=[final_label, chart_output])