Update module

app.py

@@ -83,19 +83,19 @@ avse_model.eval()
 @spaces.GPU
 def run_avse_inference(video_path, audio_path):
     # Load audio
-    noisy, _ = sf.read(audio_path, dtype='float32')  # (N, )
-    noisy = torch.tensor(noisy).unsqueeze(0)  # (1, N)
+    #noisy, _ = sf.read(audio_path, dtype='float32')  # (N, )
+    #noisy = torch.tensor(noisy).unsqueeze(0)  # (1, N)
+    noisy = wavfile.read(audio_path)[1].astype(np.float32) / (2 ** 15)
 
     # Load grayscale video
     vr = VideoReader(video_path, ctx=cpu(0))
     frames = vr.get_batch(list(range(len(vr)))).asnumpy()
     bg_frames = np.array([cv2.cvtColor(f, cv2.COLOR_RGB2GRAY) for f in frames]).astype(np.float32) / 255.0
-    bg_frames = torch.tensor(bg_frames).unsqueeze(0).unsqueeze(0)  # (1, 1, T, H, W)
 
     # Combine into input dict (match what model.enhance expects)
     data = {
-        "noisy_audio": noisy.numpy(),
-        "video_frames": bg_frames.numpy()
+        "noisy_audio": noisy,
+        "video_frames": bg_frames
     }
 
     with torch.no_grad():
@@ -189,7 +189,6 @@ def extract_faces(video_file):
 
     # ------------------------------- #
     # AVSE models
-    noisy = wavfile.read(audio_path)[1].astype(np.float32) / (2 ** 15)
 
     vr = VideoReader(output_path, ctx=cpu(0))
     frames = vr.get_batch(list(range(len(vr)))).asnumpy()