Fixed bug + added comments

src/main.py

@@ -191,6 +191,7 @@ def preprocess_song(song_input, mdx_model_params, song_id, is_webui, input_type,
 
     return orig_song_path, vocals_path, instrumentals_path, main_vocals_path, backup_vocals_path, main_vocals_dereverb_path
 
+# Function to preprocess vocals only, i.e. apply just dereverb process.
 def preprocess_vocals_only(song_input, mdx_model_params, song_id, is_webui, input_type, progress=None):
     orig_song_path = song_input
 
@@ -217,6 +218,7 @@ 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):
+    # Added _covervocals at the end of filename
     output_path = f'{os.path.splitext(audio_path)[0]}_mixed_covervocals.wav'
 
     # Initialize audio effects plugins
@@ -245,6 +247,8 @@ def combine_audio(audio_paths, output_path, main_gain, backup_gain, inst_gain, o
     instrumental_audio = AudioSegment.from_wav(audio_paths[2]) - 7 + inst_gain
     main_vocal_audio.overlay(backup_vocal_audio).overlay(instrumental_audio).export(output_path, format=output_format)
 
+# Function defining the pipeline to create the AI cover of an audio containing only voice information.
+# Returns the path of the cover vocal.
 def vocal_only_pipeline(song_input, voice_model, pitch_change, is_webui=0,index_rate=0.5, filter_radius=3,
                         rms_mix_rate=0.25, f0_method='rmvpe', crepe_hop_length=128, protect=0.33, pitch_change_all=0,
                         reverb_rm_size=0.15, reverb_wet=0.2, reverb_dry=0.8, reverb_damping=0.7, output_format='mp3',progress=gr.Progress()):
@@ -266,7 +270,19 @@ def vocal_only_pipeline(song_input, voice_model, pitch_change, is_webui=0,index_
         song_dir = os.path.join(output_dir, song_id)
         os.makedirs(song_dir)
         
-        orig_song_path, vocals_path = preprocess_vocals_only(song_input, mdx_model_params, song_id, is_webui, input_type, progress)
+        if not os.path.exists(song_dir):
+                os.makedirs(song_dir)
+                orig_song_path, vocals_path = preprocess_vocals_only(song_input, mdx_model_params, song_id, is_webui, input_type, progress)
+        else:
+            vocals_path = None
+            paths = get_audio_paths(song_dir)
+
+            # if any of the audio files aren't available or keep intermediate files, rerun preprocess
+            if any(path is None for path in paths):
+                orig_song_path, vocals_path= preprocess_vocals_only(song_input, mdx_model_params, song_id, is_webui, input_type, progress)
+            else:
+                orig_song_path = paths
+        #orig_song_path, vocals_path = preprocess_vocals_only(song_input, mdx_model_params, song_id, is_webui, input_type, progress)
         
         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')
@@ -336,6 +352,7 @@ 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')
+            # Added _cover at the end of filename
             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}')
 
             if not os.path.exists(ai_vocals_path):
@@ -362,19 +379,6 @@ def song_cover_pipeline(song_input, voice_model, pitch_change, keep_files,
                     if file and os.path.exists(file):
                         os.remove(file)
 
-            # Add _stemname to each stem
-            #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

@@ -238,7 +238,8 @@ class MDX:
         assert len(processed_batches) == len(waves), 'Incomplete processed batches, please reduce batch size!'
         return self.segment(processed_batches, True, chunk)
 
-
+# Added _stemname1 and parameters to modify the path to add "_origvocals" and "_originstr",
+# see preprocess_song() in main.py
 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=""):
     device = torch.device('cuda:0') if torch.cuda.is_available() else torch.device('cpu')
 

src/webui.py

@@ -245,6 +245,8 @@ if __name__ == '__main__':
                                      protect, f0_method, crepe_hop_length, pitch_all, reverb_rm_size, reverb_wet,
                                      reverb_dry, reverb_damping, output_format, ai_cover])
             
+            gr.Label('If you want to modify just voice content...', show_label=False)
+
             with gr.Row():
                 with gr.Column():
                     rvc_model = gr.Dropdown(voice_models, label='Voice Models', info='Models folder "AICoverGen --> rvc_models". After new models are added into this folder, click the refresh button')
@@ -254,10 +256,6 @@ if __name__ == '__main__':
                     local_file = gr.File(label='Audio file')
                     song_input_file = gr.UploadButton('Upload 📂', file_types=['audio'], variant='primary')
                     song_input_file.upload(process_file_upload, inputs=[song_input_file], outputs=[local_file, song_input])
-
-                with gr.Column():
-                    pitch = gr.Slider(-3, 3, value=0, step=1, label='Pitch Change (Vocals ONLY)', info='Generally, use 1 for male to female conversions and -1 for vice-versa. (Octaves)')
-                    pitch_all = gr.Slider(-12, 12, value=0, step=1, label='Overall Pitch Change', info='Changes pitch/key of vocals and instrumentals together. Altering this slightly reduces sound quality. (Semitones)')
             
             with gr.Row():
                 ai_vocals =gr.Audio(label='ai_vocals', show_share_button=False)
@@ -268,10 +266,7 @@ if __name__ == '__main__':
                                inputs=[song_input, rvc_model, pitch, is_webui, index_rate, filter_radius, rms_mix_rate, f0_method, crepe_hop_length,
                                        protect, pitch_all, reverb_rm_size, reverb_wet, reverb_dry, reverb_damping],
                                outputs=[ai_vocals])
-            #clear_btn.click(lambda: [0, 0, 0, 0, 0.5, 3, 0.25, 0.33, 'rmvpe', 128, 0, 0.15, 0.2, 0.8, 0.7, 'mp3', None],
-            #                outputs=[pitch, main_gain, backup_gain, inst_gain, index_rate, filter_radius, rms_mix_rate,
-            #                         protect, f0_method, crepe_hop_length, pitch_all, reverb_rm_size, reverb_wet,
-            #                         reverb_dry, reverb_damping, output_format, ai_cover])
+            
 
         # Download tab
         with gr.Tab('Download model'):