Upload 2 files

src/main.py

@@ -181,7 +181,7 @@ def preprocess_song(song_input, mdx_model_params, song_id, is_webui, input_type,
     orig_song_path = convert_to_stereo(orig_song_path)
 
     display_progress('[~] Separating Vocals from Instrumental...', 0.1, is_webui, progress)
-    vocals_path, instrumentals_path = run_mdx(mdx_model_params, song_output_dir, os.path.join(mdxnet_models_dir, 'UVR-MDX-NET-Voc_FT.onnx'), orig_song_path, denoise=True, keep_orig=keep_orig, _stemname1 = "_origvocals", _stemname2 = "_originstr")
+    vocals_path, instrumentals_path = run_mdx(mdx_model_params, song_output_dir, os.path.join(mdxnet_models_dir, 'UVR-MDX-NET-Voc_FT.onnx'), orig_song_path, denoise=True, keep_orig=keep_orig, _stemname1 = "_origvocals", _stemname2="_originstr")
 
     display_progress('[~] Separating Main Vocals from Backup Vocals...', 0.2, is_webui, progress)
     backup_vocals_path, main_vocals_path = run_mdx(mdx_model_params, song_output_dir, os.path.join(mdxnet_models_dir, 'UVR_MDXNET_KARA_2.onnx'), vocals_path, suffix='Backup', invert_suffix='Main', denoise=True)

src/mdx.py

@@ -65,9 +65,11 @@ class MDX:
     def __init__(self, model_path: str, params: MDXModel, processor=DEFAULT_PROCESSOR):
 
         # Set the device and the provider (CPU or CUDA)
-        self.device = torch.device(f'cuda:{processor}') if processor >= 0 else torch.device('cpu')
-        self.provider = ['CUDAExecutionProvider'] if processor >= 0 else ['CPUExecutionProvider']
-
+        #self.device = torch.device(f'cuda:{processor}') if processor >= 0 else torch.device('cpu')
+        self.device = torch.device('cuda:0') if torch.cuda.is_available() else torch.device('cpu')
+        #self.provider = ['CUDAExecutionProvider'] if processor >= 0 else ['CPUExecutionProvider']
+        self.provider = ['CPUExecutionProvider']
+        
         self.model = params
 
         # Load the ONNX model using ONNX Runtime
@@ -166,6 +168,8 @@ class MDX:
             waves = np.array(wave_p[:, i:i + self.model.chunk_size])
             mix_waves.append(waves)
 
+        print(self.device)
+
         mix_waves = torch.tensor(mix_waves, dtype=torch.float32).to(self.device)
 
         return mix_waves, pad, trim
@@ -285,3 +289,4 @@ def run_mdx(model_params, output_dir, model_path, filename, exclude_main=False,
     del mdx_sess, wave_processed, wave
     gc.collect()
     return main_filepath, invert_filepath
+