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)
 
     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)
@@ -217,7 +217,7 @@ def voice_change(voice_model, vocals_path, output_path, pitch_change, f0_method,
 
 
 def add_audio_effects(audio_path, reverb_rm_size, reverb_wet, reverb_dry, reverb_damping):
-    output_path = f'{os.path.splitext(audio_path)[0]}_mixed_covervocals.wav'
+    output_path = f'{os.path.splitext(audio_path)[0]}_mixed.wav'
 
     # Initialize audio effects plugins
     board = Pedalboard(
@@ -336,7 +336,7 @@ def song_cover_pipeline(song_input, voice_model, pitch_change, keep_files,
 
             pitch_change = pitch_change * 12 + pitch_change_all
             ai_vocals_path = os.path.join(song_dir, f'{os.path.splitext(os.path.basename(orig_song_path))[0]}_{voice_model}_p{pitch_change}_i{index_rate}_fr{filter_radius}_rms{rms_mix_rate}_pro{protect}_{f0_method}{"" if f0_method != "mangio-crepe" else f"_{crepe_hop_length}"}.wav')
-            ai_cover_path = os.path.join(song_dir, f'{os.path.splitext(os.path.basename(orig_song_path))[0]} ({voice_model} Ver)_cover.{output_format}')
+            ai_cover_path = os.path.join(song_dir, f'{os.path.splitext(os.path.basename(orig_song_path))[0]} ({voice_model} Ver).{output_format}')
 
             if not os.path.exists(ai_vocals_path):
                 display_progress('[~] Converting voice using RVC...', 0.5, is_webui, progress)
@@ -363,14 +363,18 @@ def song_cover_pipeline(song_input, voice_model, pitch_change, keep_files,
                         os.remove(file)
 
             # Add _stemname to each stem
-            display_progress(f'[~] ai_cover_path FINAL PATH {ai_cover_path}', 0.9, is_webui, progress)
-            time.sleep(5)
-            display_progress(f'[~] vocals_path FINAL PATH {vocals_path}', 0.9, is_webui, progress)
-            time.sleep(5)
-            display_progress(f'[~] instrumentals_path FINAL PATH {instrumentals_path}', 0.9, is_webui, progress)
-            time.sleep(5)
-            display_progress(f'[~] ai_vocals_mixed_path FINAL PATH {ai_vocals_mixed_path}', 0.9, is_webui, progress)
-            time.sleep(5)
+            #ai_cover_path = add_stem_name(ai_cover_path, "_cover")
+            #display_progress(f'[~] ai_cover_path FINAL PATH {ai_cover_path}', 0.9, is_webui, progress)
+            #time.sleep(5)
+            #vocals_path = add_stem_name(vocals_path, "_origvocals")
+            #display_progress(f'[~] vocals_path FINAL PATH {vocals_path}', 0.9, is_webui, progress)
+            #time.sleep(5)
+            #instrumentals_path = add_stem_name(instrumentals_path, "_originstr")
+            #display_progress(f'[~] instrumentals_path FINAL PATH {instrumentals_path}', 0.9, is_webui, progress)
+            #time.sleep(5)
+            #ai_vocals_mixed_path = add_stem_name(ai_vocals_mixed_path, "_covervocals")
+            #display_progress(f'[~] ai_vocals_mixed_path FINAL PATH {ai_vocals_mixed_path}', 0.9, is_webui, progress)
+            #time.sleep(5)
 
             # Returning the stems: AI cover, original vocal, original instrumental, AI generated vocal
 

src/mdx.py

@@ -239,11 +239,12 @@ class MDX:
         return self.segment(processed_batches, True, chunk)
 
 
-def run_mdx(model_params, output_dir, model_path, filename, exclude_main=False, exclude_inversion=False, suffix=None, invert_suffix=None, denoise=False, keep_orig=True, m_threads=2, _stemname1="", _stemname2=""):
+def run_mdx(model_params, output_dir, model_path, filename, exclude_main=False, exclude_inversion=False, suffix=None, invert_suffix=None, denoise=False, keep_orig=True, m_threads=2):
     device = torch.device('cuda:0') if torch.cuda.is_available() else torch.device('cpu')
 
-    device_properties = torch.cuda.get_device_properties(device)
-    vram_gb = device_properties.total_memory / 1024**3
+    #device_properties = torch.cuda.get_device_properties(device)
+    print("Device", device)
+    vram_gb = 12 #device_properties.total_memory / 1024**3
     m_threads = 1 if vram_gb < 8 else 2
 
     model_hash = MDX.get_hash(model_path)
@@ -273,14 +274,14 @@ def run_mdx(model_params, output_dir, model_path, filename, exclude_main=False,
 
     main_filepath = None
     if not exclude_main:
-        main_filepath = os.path.join(output_dir, f"{os.path.basename(os.path.splitext(filename)[0])}_{stem_name}_{_stemname1}.wav")
+        main_filepath = os.path.join(output_dir, f"{os.path.basename(os.path.splitext(filename)[0])}_{stem_name}.wav")
         sf.write(main_filepath, wave_processed.T, sr)
 
     invert_filepath = None
     if not exclude_inversion:
         diff_stem_name = stem_naming.get(stem_name) if invert_suffix is None else invert_suffix
         stem_name = f"{stem_name}_diff" if diff_stem_name is None else diff_stem_name
-        invert_filepath = os.path.join(output_dir, f"{os.path.basename(os.path.splitext(filename)[0])}_{stem_name}_{_stemname2}.wav")
+        invert_filepath = os.path.join(output_dir, f"{os.path.basename(os.path.splitext(filename)[0])}_{stem_name}.wav")
         sf.write(invert_filepath, (-wave_processed.T * model.compensation) + wave.T, sr)
 
     if not keep_orig:
@@ -289,4 +290,3 @@ 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
-