vocal_pipiline

src/main.py

@@ -189,6 +189,22 @@ 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
 
+def preprocess_vocals_only(song_input, mdx_model_params, song_id, is_webui, input_type, progress=None):
+    keep_orig = False
+    if input_type == 'local':
+        orig_song_path = song_input
+        keep_orig = True
+    else:
+        orig_song_path = None
+
+    song_output_dir = os.path.join(output_dir, song_id)
+    orig_song_path = convert_to_stereo(orig_song_path)
+
+    display_progress('[~] Applying DeReverb to Vocals...', 0.3, is_webui, progress)
+    _, vocals_path = run_mdx(mdx_model_params, song_output_dir, os.path.join(mdxnet_models_dir, 'Reverb_HQ_By_FoxJoy.onnx'), orig_song_path, invert_suffix='DeReverb', exclude_main=True, denoise=True)
+    
+    return vocals_path
+
 
 def voice_change(voice_model, vocals_path, output_path, pitch_change, f0_method, index_rate, filter_radius, rms_mix_rate, protect, crepe_hop_length, is_webui):
     rvc_model_path, rvc_index_path = get_rvc_model(voice_model, is_webui)
@@ -232,6 +248,49 @@ 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)
 
+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()):
+    #Load mdx parameters
+    with open(os.path.join(mdxnet_models_dir, 'model_data.json')) as infile:
+        mdx_model_params = json.load(infile)
+
+    # Get path of loaded vocal file
+    input_type = 'local'
+    song_input = song_input.strip('\"')
+    if os.path.exists(song_input):
+        song_id = get_hash(song_input)
+    else:
+        error_msg = f'{song_input} does not exist.'
+        song_id = None
+        raise_exception(error_msg, is_webui)
+        
+    song_dir = os.path.join(output_dir, song_id)
+
+    if not os.path.exists(song_dir):
+        os.makedirs(song_dir)
+        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):
+            vocals_path = preprocess_vocals_only(song_input, mdx_model_params, song_id, is_webui, input_type, progress)
+        else:
+            vocals_path = paths
+        
+        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')
+        if not os.path.exists(ai_vocals_path):
+            display_progress('[~] Converting voice using RVC...', 0.5, is_webui, progress)
+            voice_change(voice_model, vocals_path, ai_vocals_path, pitch_change, f0_method, index_rate, filter_radius, rms_mix_rate, protect, crepe_hop_length, is_webui)
+        display_progress('[~] Applying audio effects to Vocals...', 0.8, is_webui, progress)
+        ai_vocals_mixed_path = add_audio_effects(ai_vocals_path, reverb_rm_size, reverb_wet, reverb_dry, reverb_damping)
+
+        return ai_vocals_mixed_path
+
+
 
 def song_cover_pipeline(song_input, voice_model, pitch_change, keep_files,
                         is_webui=0, main_gain=0, backup_gain=0, inst_gain=0, index_rate=0.5, filter_radius=3,
@@ -355,8 +414,10 @@ if __name__ == '__main__':
                                      pitch_change_all=args.pitch_change_all,
                                      reverb_rm_size=args.reverb_size, reverb_wet=args.reverb_wetness,
                                      reverb_dry=args.reverb_dryness, reverb_damping=args.reverb_damping,
-                                     output_format=args.output_format)
+                                     output_format=args.output_format, vocal_only=False)
     
+    
+
     print(f'[+] Cover generated at {cover_path}')
     print(f'[+] Original vocals at {original_vocals}')
     print(f'[+] Original instrumentals at {original_instrumentals}')

src/webui.py

@@ -7,7 +7,7 @@ from argparse import ArgumentParser
 
 import gradio as gr
 
-from main import song_cover_pipeline
+from main import song_cover_pipeline, vocal_only_pipeline
 
 BASE_DIR = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
 
@@ -181,6 +181,7 @@ if __name__ == '__main__':
 
                     with gr.Column() as yt_link_col:
                         song_input = gr.Text(label='Song input', info='Link to a song on YouTube or full path to a local file. For file upload, click the button below.')
+                        
                         show_file_upload_button = gr.Button('Upload file instead')
 
                     with gr.Column(visible=False) as file_upload_col:
@@ -243,6 +244,33 @@ if __name__ == '__main__':
                             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])
+            
+            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')
+                    ref_btn = gr.Button('Refresh Models 🔁', variant='primary')
+
+                with gr.Column() as file_upload_col:
+                    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)
+            ref_btn.click(update_models_list, None, outputs=rvc_model)
+            is_webui = gr.Number(value=1, visible=False)
+            generate_btn.click(vocal_only_pipeline,
+                               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'):