Update app.py

app.py

@@ -1,51 +1,40 @@
 import gradio as gr
-import tensorflow as tf
 import librosa
 import numpy as np
+import tensorflow as tf
+from tensorflow.keras.models import load_model
 
-# Load the saved model
-model = tf.keras.models.load_model("voice_authentication_model.keras")
-
-# Function to extract features from audio
-def extract_features(file_path):
-    try:
-        # Load the audio file
-        audio, sample_rate = librosa.load(file_path, res_type='kaiser_fast', duration=3, sr=None)
-        
-        # Extract MFCC features (13 MFCC coefficients)
-        mfcc = librosa.feature.mfcc(y=audio, sr=sample_rate, n_mfcc=13)
-        mfcc = np.mean(mfcc.T, axis=0)  # Taking the mean over time
-        
-        return mfcc
-    except Exception as e:
-        print(f"Error encountered while parsing file: {file_path}. Error: {e}")
-        return None
+# Load your pre-trained model (make sure it's in the same directory or provide a path)
+model = load_model('voice_authentication_model.keras')  # Replace with your model's filename
 
-# Prediction function for user vs non-user
-def predict_user_or_nonuser(audio):
-    # Extract features from the uploaded audio
-    feature = extract_features(audio.name)  # Audio file is passed as Gradio interface's input
-    if feature is None:
-        return "Error: Unable to process audio"
+# Function to extract MFCC features and make a prediction
+def predict_user_or_non_user(audio):
+    # Load the audio file using librosa
+    y, sr = librosa.load(audio, sr=None)  # sr=None to keep the original sampling rate
+    
+    # Extract MFCC features from the audio
+    mfccs = librosa.feature.mfcc(y=y, sr=sr, n_mfcc=13)  # You can adjust n_mfcc as needed
+    mfccs = np.mean(mfccs.T, axis=0)  # Take the mean of MFCCs over time to reduce dimension
     
-    # Reshape feature to match the model's input format (1, time_steps, 1)
-    feature = feature.reshape(1, feature.shape[0], 1)
+    # Reshape the MFCCs to match the input shape expected by the model
+    mfccs = mfccs.reshape(1, -1)  # Reshape to 1 sample, with the number of features
     
-    # Make prediction
-    prediction = model.predict(feature)
+    # Predict the class (user or non-user)
+    prediction = model.predict(mfccs)
     
-    # Return result: User or Non-User
-    return "User" if prediction > 0.5 else "Non-User"
+    # Convert prediction to readable format
+    if prediction > 0.5:
+        return "Non-user"
+    else:
+        return "User"
 
-# Create a Gradio interface for the app
+# Define the Gradio interface
 iface = gr.Interface(
-    fn=predict_user_or_nonuser,  # Function to call for prediction
-    inputs=gr.Audio(source="upload", type="file"),  # Audio input
-    outputs="text",  # Output: Text label ("User" or "Non-User")
-    title="Voice Authentication",
-    description="This application can authenticate users based on their voice. Upload an audio file to check if it's from the user or a non-user."
+    fn=predict_user_or_non_user,  # The function to call when an audio input is given
+    inputs=gr.Audio(source="microphone", type="filepath"),  # Audio input from the microphone
+    outputs="text",  # Output will be text (User or Non-user)
+    live=True  # Live mode so the interface updates in real-time
 )
 
-# Launch the Gradio app
-if __name__ == "__main__":
-    iface.launch()
+# Launch the interface
+iface.launch()