Added files

app.py

@@ -3,9 +3,10 @@ import torch
 import torchaudio
 import numpy as np
 import plotly.graph_objs as go
-import os  # Import os for file operations
-from model import BoundaryDetectionModel  # Assuming your model is defined here
-from audio_dataset import pad_audio  # Assuming you have a function to pad audio
+import os  # For file operations
+from pydub import AudioSegment  # For audio format conversion
+from model import BoundaryDetectionModel  
+from audio_dataset import pad_audio  
 
 app = Flask(__name__)
 
@@ -15,7 +16,17 @@ model = BoundaryDetectionModel().to(device)
 model.load_state_dict(torch.load("checkpoint_epoch_21_eer_0.24.pth", map_location=device)["model_state_dict"])
 model.eval()
 
+def convert_to_wav(audio_path, temp_path="temp_audio.wav"):
+    # Check if the file is already in .wav format
+    if audio_path.lower().endswith(".wav"):
+        return audio_path
+    # Convert to .wav using pydub if it's not already in .wav
+    audio = AudioSegment.from_file(audio_path)
+    audio.export(temp_path, format="wav")
+    return temp_path
+
 def preprocess_audio(audio_path, sample_rate=16000, target_length=8):
+    # Load the audio waveform
     waveform, sr = torchaudio.load(audio_path)
     if sr != sample_rate:
         waveform = torchaudio.transforms.Resample(sr, sample_rate)(waveform)
@@ -41,8 +52,10 @@ def predict():
     if file.filename == '':
         return "No selected file", 400
 
-    file_path = "temp_audio.wav"  # Temporary file to store uploaded audio
-    file.save(file_path)
+    # Save the file to a temporary location and convert if necessary
+    original_path = "temp_uploaded_audio"
+    file.save(original_path)
+    file_path = convert_to_wav(original_path)  # Convert to .wav if needed
     
     # Preprocess audio and perform inference
     audio_tensor = preprocess_audio(file_path)
@@ -83,9 +96,8 @@ def predict():
 def return_to_index():
     # Delete temporary files before returning to index
     try:
-        os.remove("temp_audio.wav")  # Remove the temporary audio file
-        # If you have any other temporary files (like plots), remove them here too.
-        # Example: os.remove("temp_plot.html") if you save plots as HTML files.
+        os.remove("temp_uploaded_audio")  # Remove original uploaded audio file
+        os.remove("temp_audio.wav")  # Remove the converted .wav file if necessary
     except OSError as e:
         print(f"Error deleting temporary files: {e}")
     
@@ -157,4 +169,4 @@ def plot_fake_frames_waveform(output, prediction_flat, waveform, fake_frame_inte
     return plot_html
 
 if __name__ == '__main__':
-    app.run()
+    app.run()

requirements.txt

@@ -4,4 +4,5 @@ librosa
 flask
 gunicorn
 uvicorn
-plotly
+plotly
+pydub