Update app.py

app.py

@@ -1,9 +1,21 @@
 import os
 import subprocess
 import sys
+import fileinput
+import gradio as gr
 
-# --- 1. Clone the VibeVoice Repository ---
-# Check if the repository directory already exists
+# --- 1. Setup Environment ---
+# Install ffmpeg if it's not already installed by packages.txt
+print("Updating apt and installing ffmpeg...")
+try:
+    subprocess.run(["apt-get", "update", "-y"], check=True)
+    subprocess.run(["apt-get", "install", "ffmpeg", "-y"], check=True)
+    print("ffmpeg installed successfully.")
+except subprocess.CalledProcessError as e:
+    print(f"Error installing ffmpeg: {e.stderr}")
+    # Continue anyway, as it might be pre-installed
+
+# --- 2. Clone the VibeVoice Repository ---
 repo_dir = "VibeVoice"
 if not os.path.exists(repo_dir):
     print("Cloning the VibeVoice repository...")
@@ -21,14 +33,11 @@ if not os.path.exists(repo_dir):
 else:
     print("Repository already exists. Skipping clone.")
 
-# --- 2. Install the Package ---
-# Navigate into the repository directory
+# --- 3. Install the Package ---
 os.chdir(repo_dir)
 print(f"Changed directory to: {os.getcwd()}")
-
 print("Installing the VibeVoice package...")
 try:
-    # Use pip to install the package in editable mode
     subprocess.run(
         [sys.executable, "-m", "pip", "install", "-e", "."],
         check=True,
@@ -40,13 +49,35 @@ except subprocess.CalledProcessError as e:
     print(f"Error installing package: {e.stderr}")
     sys.exit(1)
 
-# --- 3. Launch the Gradio Demo ---
-# Define the path to the demo script and the model to use
+
+# --- 4. Modify the demo script for CPU execution ---
 demo_script_path = "demo/gradio_demo.py"
+print(f"Modifying {demo_script_path} for CPU execution...")
+
+# Use fileinput to perform in-place replacement
+with fileinput.FileInput(demo_script_path, inplace=True) as file:
+    for line in file:
+        # Change the model loading line to be CPU-compatible
+        if 'VibeVoiceForConditionalGenerationInference.from_pretrained(' in line:
+            print('        self.model = VibeVoiceForConditionalGenerationInference.from_pretrained(')
+            print('            self.model_path,')
+            print('            torch_dtype=torch.float32,  # Use float32 for CPU')
+            print('            device_map="cpu",')
+            print('        )')
+            # Skip the next few lines of the original arguments
+            next(file, None)
+            next(file, None)
+            next(file, None)
+        else:
+            print(line, end='')
+
+print("Script modified successfully.")
+
+
+# --- 5. Launch the Gradio Demo ---
 model_id = "microsoft/VibeVoice-1.5B"
 
-# Construct the command to run the demo
-# The --share flag is necessary to make the Gradio app accessible within the Hugging Face Space environment
+# Construct the command as specified in the README
 command = [
     "python",
     demo_script_path,
@@ -56,6 +87,5 @@ command = [
 ]
 
 print(f"Launching Gradio demo with command: {' '.join(command)}")
-# Run the command. This will start the Gradio server and launch the demo.
-# The process will remain active, serving the web interface.
+# This command will start the Gradio server
 subprocess.run(command)