Update app.py

app.py

@@ -227,6 +227,8 @@ def load_input_spectrogram(audio_path):
     input_spec_image_path = "input_spectrogram.png"
     raw_image.save(input_spec_image_path)
 
+    return input_spec_image_path
+
 def preview_masked_area(audio_path, mask_start_point, mask_end_point):
     # Loading
     audio, sampling_rate = load_wav(audio_path)
@@ -236,6 +238,7 @@ def preview_masked_area(audio_path, mask_start_point, mask_end_point):
     norm_spec = normalize(norm_spec) # normalize to [-1, 1], because pipeline do not normalize for torch.Tensor input
 
     # Add Mask
+    width_start, width = mask_start_point, mask_end_point-mask_start_point
     mask = torch.zeros_like(norm_spec)[:1,...]
     mask[:, :, width_start:width_start+width] = 1
     mask_image = torch_to_pil(mask)