Update app.py

app.py

@@ -3,6 +3,7 @@ import json
 from flask import Flask, jsonify, request
 from transformers import pipeline
 from pydub import AudioSegment
+from io import BytesIO
 
 # Create a Flask app
 app = Flask(__name__)
@@ -13,17 +14,32 @@ audio_model = None
 def download_models():
     global audio_model
     print("Downloading models...")
-    # Download and load the audio model with padding enabled
-    audio_model = pipeline("audio-classification", model="MelodyMachine/Deepfake-audio-detection-V2", padding=True)
+    # Download and load the audio model
+    audio_model = pipeline("audio-classification", model="MelodyMachine/Deepfake-audio-detection-V2")
     print("Model downloaded and ready to use.")
 
 # Download model when the server starts
 download_models()
 
-def convert_audio_to_wav(input_path, output_path):
-    # Convert any audio format to WAV using pydub
-    audio = AudioSegment.from_file(input_path)
-    audio.export(output_path, format="wav")
+def preprocess_audio(file):
+    # Load audio file
+    audio = AudioSegment.from_file(file)
+    
+    # Convert audio to mono and normalize volume
+    audio = audio.set_channels(1).set_frame_rate(16000)
+    
+    # Ensure audio is of a standard length (e.g., 10 seconds)
+    duration_ms = len(audio)
+    target_duration_ms = 10000  # Target duration in milliseconds (10 seconds)
+    if duration_ms < target_duration_ms:
+        # Pad with silence if shorter than target duration
+        padding = AudioSegment.silent(duration=target_duration_ms - duration_ms)
+        audio = audio + padding
+    elif duration_ms > target_duration_ms:
+        # Truncate if longer than target duration
+        audio = audio[:target_duration_ms]
+
+    return audio
 
 @app.route('/detect', methods=['POST'])
 def detect_deepfake():
@@ -33,22 +49,18 @@ def detect_deepfake():
     # If a single audio file is provided
     if audio_file:
         try:
-            # Save the uploaded file temporarily
-            input_path = os.path.join("/tmp", audio_file.filename)
-            audio_file.save(input_path)
-
-            # Convert the file to WAV format
-            output_path = os.path.splitext(input_path)[0] + '.wav'
-            convert_audio_to_wav(input_path, output_path)
+            # Preprocess the audio file
+            audio = preprocess_audio(audio_file)
+            
+            # Save the processed file temporarily
+            temp_wav = BytesIO()
+            audio.export(temp_wav, format="wav")
+            temp_wav.seek(0)
 
             # Perform detection
-            result = audio_model(output_path)
+            result = audio_model(temp_wav)
             result_dict = {item['label']: item['score'] for item in result}
 
-            # Remove the temporary files
-            os.remove(input_path)
-            os.remove(output_path)
-
             return jsonify({"message": "Detection completed", "results": result_dict}), 200
 
         except Exception as e: