Update app.py

app.py

@@ -155,40 +155,45 @@ def analyze_audio(audio):
     try:
         if audio is None:
             return "🎤 No audio", "0%", "Please record/upload audio.", None
-        
+
         sample_rate, audio_data = audio
-        
-        # Enhanced preprocessing
+
+        if audio_data is None or len(audio_data) == 0:
+            return "🎤 Invalid audio", "0%", "Unreadable audio.", None
+
+        # Stereo → mono
         if len(audio_data.shape) > 1:
-            audio_data = np.mean(audio_data, axis=1)  # Stereo to mono
-        
-        # Resample to 16kHz if needed (optimal for Wav2Vec2)
-        if sample_rate != 16000:
-            audio_data = librosa.resample(
-                audio_data, 
-                orig_sr=sample_rate, 
-                target_sr=16000
-            )
-            sample_rate = 16000
-        
-        # Better normalization
-        if audio_data.dtype != np.float32:
-            audio_data = audio_data.astype(np.float32)
-            if np.abs(audio_data).max() > 0:
-                audio_data = audio_data / np.abs(audio_data).max()
-        
-        # Consider using the more accurate model
-        # model_name = "r-f/wav2vec-english-speech-emotion-recognition"
-        # pipe = pipeline(
-        #     "audio-classification",
-        #     model=model_name,
-        #     device=-1  # Use -1 for CPU, 0 for GPU if available
-        # )
-        
-        preds = pipe({"raw": audio_data, "sampling_rate": int(sample_rate)})
-        
-        # Rest of your function...
-        
+            audio_data = np.mean(audio_data, axis=1)
+
+        # Normalize
+        audio_data = audio_data.astype(np.float32)
+        max_val = np.abs(audio_data).max()
+        if max_val > 0:
+            audio_data /= max_val
+
+        preds = pipe({
+            "raw": audio_data,
+            "sampling_rate": int(sample_rate)
+        })
+
+        top = preds[0]
+        label = top["label"].upper()
+        confidence = f"{top['score'] * 100:.1f}%"
+
+        emoji_map = {
+            "ANGER": "😠", "DISGUST": "🤢", "FEAR": "😨",
+            "HAPPY": "😊", "NEUTRAL": "😐",
+            "SADNESS": "😢", "SURPRISE": "😲"
+        }
+
+        mood_display = emoji_map.get(label, label)
+        details = "\n".join(
+            f"{p['label'].upper()}: {p['score'] * 100:.1f}%"
+            for p in preds[:6]
+        )
+
+        return mood_display, confidence, details, label
+
     except Exception as e:
         return "❌ Error", "0%", f"Analysis failed: {str(e)}", None