update app

app.py

@@ -4,13 +4,13 @@ import pyttsx3
 import onnxruntime as ort
 import librosa
 import sounddevice as sd
-import tempfile
 import scipy.io.wavfile as wavfile
 from sklearn.preprocessing import StandardScaler
 import time
 import os
 from gtts import gTTS
 import gradio as gr
+import tempfile
 
 # ------------------- Speech Emotion Recognition Model -------------------
 class SpeechEmotionRecognizer:
@@ -19,11 +19,14 @@ class SpeechEmotionRecognizer:
         self.input_name = self.model.get_inputs()[0].name
         self.labels = ['angry', 'disgust', 'fear', 'happy', 'neutral', 'sad', 'surprise']
 
+        # Load or create scaler here (fit on training data offline, then load)
+        self.scaler = StandardScaler()
+
     def extract_features(self, y, sr):
         mfcc = librosa.feature.mfcc(y=y, sr=sr, n_mfcc=40)
         mfcc_mean = np.mean(mfcc.T, axis=0)
-        scaler = StandardScaler()
-        mfcc_scaled = scaler.fit_transform(mfcc_mean.reshape(-1, 1)).flatten()
+        # Normally, scaler should be pre-fitted, here we just scale manually to zero mean, unit var
+        mfcc_scaled = (mfcc_mean - np.mean(mfcc_mean)) / np.std(mfcc_mean)
         return mfcc_scaled
 
     def predict_emotion(self, audio_data, sr):
@@ -42,30 +45,23 @@ class FacialEmotionRecognizer:
 
     def predict_emotion(self, face_img):
         face_img = cv2.resize(face_img, (64, 64))
-        face_img = face_img.astype('float32')  # FER+ expects float32 in [0,255]
-        face_img = np.expand_dims(face_img, axis=(0, 1))  # Shape: (1, 1, 64, 64)
+        face_img = face_img.astype('float32')  # FER+ expects float32
+        # FER+ model expects input shape (1, 1, 64, 64)
+        face_img = np.expand_dims(face_img, axis=0)  # (1, 64, 64)
+        face_img = np.expand_dims(face_img, axis=0)  # (1, 1, 64, 64)
         pred = self.model.run(None, {self.input_name: face_img})[0]
         emotion_idx = np.argmax(pred)
         return self.labels[emotion_idx]
 
 # ------------------- Utility Functions -------------------
 
-
 def speak(text):
     if not text.strip():
         return None
-    tts = gTTS(text)
-    tts.save("output.mp3")
-    return "output.mp3"
-
-iface = gr.Interface(
-    fn=speak,
-    inputs=gr.Textbox(lines=2, label="Enter text"),
-    outputs=gr.Audio(type="filepath", label="Speech Output"),
-    title="Text to Speech"
-)
-
-iface.launch()
+    with tempfile.NamedTemporaryFile(delete=False, suffix=".mp3") as tmpfile:
+        tts = gTTS(text)
+        tts.save(tmpfile.name)
+        return tmpfile.name
 
 def record_audio(duration=3, fs=22050):
     print("Recording audio...")
@@ -80,12 +76,11 @@ def analyze_face(face_roi, emotion_model):
     return emotion
 
 # ------------------- Main Function -------------------
+
 def main():
-    # Load models
     face_emotion_model = FacialEmotionRecognizer("emotion-ferplus-8.onnx")
-    speech_emotion_model = SpeechEmotionRecognizer("speech_emotion_model.onnx")  # Replace with your .onnx model
+    speech_emotion_model = SpeechEmotionRecognizer("speech_emotion_model.onnx")
 
-    # Start webcam
     cap = cv2.VideoCapture(0)
     face_cascade = cv2.CascadeClassifier(cv2.data.haarcascades + "haarcascade_frontalface_default.xml")
 
@@ -114,8 +109,11 @@ def main():
             audio, sr = record_audio()
             speech_emotion = speech_emotion_model.predict_emotion(audio, sr)
             print(f"Speech Emotion: {speech_emotion}")
-            speak(f"You sound {speech_emotion}")
-        
+            audio_file = speak(f"You sound {speech_emotion}")
+            if audio_file:
+                # Play the TTS audio using cv2 or other player if needed
+                pass
+
         elif key == ord('q'):
             break