import json import os import shutil import urllib.request import zipfile from argparse import ArgumentParser import gradio as gr import logging def configure_logging_libs(debug=False): modules = [ "numba", "httpx", "markdown_it", "fairseq", "faiss", ] try: for module in modules: logging.getLogger(module).setLevel(logging.WARNING) os.environ['TF_CPP_MIN_LOG_LEVEL'] = "3" if not debug else "1" except Exception as error: pass configure_logging_libs() from main import song_cover_pipeline, yt_download BASE_DIR = os.path.dirname(os.path.dirname(os.path.abspath(__file__))) IS_ZERO_GPU = os.getenv("SPACES_ZERO_GPU") rvc_assets_dir = os.path.join(BASE_DIR, 'assets', 'rvc_models') rvc_models_dir = os.path.join(BASE_DIR, 'rvc_models') output_dir = os.path.join(BASE_DIR, 'song_output') def get_current_models(models_dir): models_list = os.listdir(models_dir) items_to_remove = ['.gitkeep'] return [item for item in models_list if item not in items_to_remove] def update_models_list(): models_l = get_current_models(rvc_models_dir) return gr.update(choices=models_l) def extract_zip(extraction_folder, zip_name): os.makedirs(extraction_folder) with zipfile.ZipFile(zip_name, 'r') as zip_ref: zip_ref.extractall(extraction_folder) os.remove(zip_name) index_filepath, model_filepath = None, None for root, dirs, files in os.walk(extraction_folder): for name in files: if name.endswith('.index') and os.stat(os.path.join(root, name)).st_size > 1024 * 100: index_filepath = os.path.join(root, name) if name.endswith('.pth') and os.stat(os.path.join(root, name)).st_size > 1024 * 1024 * 40: model_filepath = os.path.join(root, name) if not model_filepath: raise gr.Error(f'No .pth model file was found in the extracted zip. Please check {extraction_folder}.') os.rename(model_filepath, os.path.join(extraction_folder, os.path.basename(model_filepath))) if index_filepath: os.rename(index_filepath, os.path.join(extraction_folder, os.path.basename(index_filepath))) for filepath in os.listdir(extraction_folder): if os.path.isdir(os.path.join(extraction_folder, filepath)): shutil.rmtree(os.path.join(extraction_folder, filepath)) def download_online_model(url, dir_name, progress=gr.Progress()): try: progress(0, desc=f'[~] Downloading voice model with name {dir_name}...') zip_name = url.split('/')[-1] extraction_folder = os.path.join(rvc_models_dir, dir_name) if os.path.exists(extraction_folder): raise gr.Error(f'Voice model directory {dir_name} already exists! Choose a different name for your voice model.') if 'pixeldrain.com' in url: url = f'https://pixeldrain.com/api/file/{zip_name}' if "," in url: urls = [u.strip() for u in url.split(",") if u.strip()] os.makedirs(extraction_folder, exist_ok=True) for u in urls: u = u.replace("?download=true", "") file_name = u.split('/')[-1] file_path = os.path.join(extraction_folder, file_name) if not os.path.exists(file_path): urllib.request.urlretrieve(u, file_path) else: urllib.request.urlretrieve(url, zip_name) progress(0.5, desc='[~] Extracting zip...') extract_zip(extraction_folder, zip_name) return f'[+] {dir_name} Model successfully downloaded!' except Exception as e: raise gr.Error(str(e)) def upload_local_model(zip_path, dir_name, progress=gr.Progress()): try: extraction_folder = os.path.join(rvc_models_dir, dir_name) if os.path.exists(extraction_folder): raise gr.Error(f'Voice model directory {dir_name} already exists! Choose a different name for your voice model.') # Gradio 6.x with type="filepath" returns a string path, not a file object zip_name = zip_path if isinstance(zip_path, str) else zip_path.name progress(0.5, desc='[~] Extracting zip...') extract_zip(extraction_folder, zip_name) return f'[+] {dir_name} Model successfully uploaded!' except Exception as e: raise gr.Error(str(e)) def pub_dl_autofill(pub_models, event: gr.SelectData): return gr.update(value=pub_models.loc[event.index[0], 'URL']), gr.update(value=pub_models.loc[event.index[0], 'Model Name']) def show_hop_slider(pitch_detection_algo): if 'crepe' in pitch_detection_algo: return gr.update(visible=True) else: return gr.update(visible=False) def update_voice_model_visibility(mode): """Show/hide the voice model dropdown based on inference mode. Voice model is required for 'full' and 'rvc' modes, not for 'mdx' mode.""" if mode == 'mdx': return gr.update(visible=False) else: return gr.update(visible=True) def update_input_visibility(selected_method): if selected_method == "File Upload": return gr.update(visible=True), gr.update(visible=False), gr.update(visible=False) elif selected_method == "YouTube URL": return gr.update(visible=False), gr.update(visible=True), gr.update(visible=False) elif selected_method == "File Path": return gr.update(visible=False), gr.update(visible=False), gr.update(visible=True) return gr.update(visible=False), gr.update(visible=False), gr.update(visible=False) def process_file_path(file_path): if os.path.exists(file_path): return file_path, gr.update(value=f"✓ File loaded: {file_path}") else: return None, gr.update(value=f"✗ File not found: {file_path}") css = """ .title { font-size: 3em; align-items: center; text-align: center; } .info { align-items: center; text-align: center; } """ if __name__ == '__main__': parser = ArgumentParser(description='Generate a AI cover song in the song_output/id directory.', add_help=True) parser.add_argument("--share", action="store_true", dest="share_enabled", default=False, help="Enable sharing") parser.add_argument("--listen", action="store_true", default=False, help="Make the WebUI reachable from your local network.") parser.add_argument('--listen-host', type=str, help='The hostname that the server will use.') parser.add_argument('--listen-port', type=int, help='The listening port that the server will use.') parser.add_argument('--theme', type=str, default="NoCrypt/miku", help='Set the theme (default: NoCrypt/miku)') parser.add_argument("--ssr", action="store_true", help="Enable SSR (Server-Side Rendering)") args = parser.parse_args() voice_models = get_current_models(rvc_models_dir) with gr.Blocks(css=css, title='AICoverGenWebUI', theme=args.theme, fill_width=True, fill_height=True) as app: gr.Label('AICGP created with ❤️', show_label=False) # Main Generate tab with gr.Tab("Generate"): with gr.Row(equal_height=True): 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') # Inference Mode Selection with gr.Row(equal_height=True): inference_mode = gr.Radio( choices=['full', 'mdx', 'rvc'], value='full', label='Inference Mode', info='Full: MDX separation + RVC voice conversion | MDX Only: Separate vocals only (no RVC) | RVC Only: Convert voice only (no separation)', interactive=True ) # Input Method Selection with gr.Row(equal_height=True): input_method = gr.Radio( choices=["File Upload", "YouTube URL", "File Path"], label="Select Input Method", value="File Upload", interactive=True ) with gr.Column(): # File Upload Section with gr.Column(visible=True) as file_upload_col: audio_extensions = ['.mp3', '.m4a', '.flac', '.wav', '.aac', '.ogg', '.wma', '.alac', '.aiff', '.opus', '.amr'] local_file = gr.File(label='Upload Audio File', interactive=True, type="filepath", file_types=audio_extensions) # YouTube URL Section with gr.Column(visible=False) as yt_url_col: yt_file = gr.Audio(label='YT OPT', interactive=True) with gr.Column(): yt_url = gr.Textbox(label="YouTube URL", placeholder="https://www.youtube.com/watch?v=...", lines=1) process_yt_btn = gr.Button("Process YouTube URL", variant="secondary") yt_status = gr.Textbox(label="Status", interactive=False, visible=False) def process_yt_url(url): if url: try: downloaded_file = yt_download(url) return downloaded_file, gr.update(visible=True, value="✓ YouTube video processed successfully!"), gr.update(value=downloaded_file) except Exception as e: return None, gr.update(visible=True, value=f"✗ Error: {str(e)}"), gr.update(value=None) return None, gr.update(visible=True, value="✗ Please enter a valid URL"), gr.update(value=None) process_yt_btn.click(process_yt_url, inputs=[yt_url], outputs=[yt_file, yt_status, local_file]) # File Path Section with gr.Column(visible=False) as file_path_col: file_path_input = gr.Textbox(label="File Path", placeholder="/path/to/your/audio/file.mp3", lines=1) process_path_btn = gr.Button("Load File Path", variant="secondary") path_status = gr.Textbox(label="Status", interactive=False, visible=False) process_path_btn.click(process_file_path, inputs=[file_path_input], outputs=[local_file, path_status]) # Update visibility based on selection input_method.change(update_input_visibility, inputs=[input_method], outputs=[file_upload_col, yt_url_col, file_path_col]) with gr.Row(equal_height=True): 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)') # Voice conversion options with gr.Accordion('Settings', open=False): with gr.Accordion('Voice conversion options', open=False): with gr.Row(equal_height=True): index_rate = gr.Slider(0, 1, value=0.5, label='Index Rate', info="Controls how much of the AI voice's accent to keep in the vocals") filter_radius = gr.Slider(0, 7, value=3, step=1, label='Filter radius', info='If >=3: apply median filtering to the harvested pitch results. Can reduce breathiness') volume_envelope = gr.Slider(0, 1, value=0.25, label='Volume Envelope', info="Control how much to mimic the original vocal's loudness (0) or a fixed loudness (1)") protect = gr.Slider(0, 0.5, value=0.33, label='Protect rate', info='Protect voiceless consonants and breath sounds. Set to 0.5 to disable.') with gr.Column(): f0_method = gr.Dropdown( ['rmvpe', 'rmvpe-legacy', 'mangio-crepe', 'mangio-crepe-tiny', 'mangio-crepe-small', 'mangio-crepe-medium', 'mangio-crepe-large', 'mangio-crepe-full', 'crepe-tiny', 'crepe-small', 'crepe-medium', 'crepe-large', 'crepe-full', 'fcpe', 'fcpe-legacy', 'djcm', 'harvest', 'yin', 'pyin', 'swipe', 'dio', 'pm'], value='rmvpe', label='Pitch detection algorithm', info='Best: rmvpe (clarity), mangio-crepe variants (smoother), fcpe (fast). Legacy versions available for compatibility.' ) hop_length = gr.Slider(32, 320, value=128, step=1, visible=False, label='Hop Length', info='Lower values leads to longer conversions and higher risk of voice cracks, but better pitch accuracy.') f0_method.change(show_hop_slider, inputs=f0_method, outputs=hop_length) with gr.Row(equal_height=True): extra_denoise = gr.Checkbox(True, label='Denoise', info='Apply an additional noise reduction step to clean up the audio further.') keep_files = gr.Checkbox((False if IS_ZERO_GPU else True), label='Keep intermediate files', info='Keep all audio files generated in the song_output/id directory, e.g. Isolated Vocals/Instrumentals. Leave unchecked to save space', interactive=(False if IS_ZERO_GPU else True)) # Audio mixing options with gr.Accordion('Audio mixing options', open=False): gr.Markdown('### Volume Change (decibels)') with gr.Row(): main_gain = gr.Slider(-20, 20, value=0, step=1, label='Main Vocals') backup_gain = gr.Slider(-20, 20, value=0, step=1, label='Backup Vocals') inst_gain = gr.Slider(-20, 20, value=0, step=1, label='Music') gr.Markdown('### Reverb Control on AI Vocals') with gr.Row(): reverb_rm_size = gr.Slider(0, 1, value=0.15, label='Room size', info='The larger the room, the longer the reverb time') reverb_wet = gr.Slider(0, 1, value=0.2, label='Wetness level', info='Level of AI vocals with reverb') reverb_dry = gr.Slider(0, 1, value=0.8, label='Dryness level', info='Level of AI vocals without reverb') reverb_damping = gr.Slider(0, 1, value=0.7, label='Damping level', info='Absorption of high frequencies in the reverb') gr.Markdown('### Audio Output Format') output_format = gr.Dropdown(['mp3', 'wav'], value='mp3', label='Output file type', info='mp3: small file size, decent quality. wav: Large file size, best quality') with gr.Row(equal_height=True): clear_btn = gr.ClearButton(value='Clear', components=[local_file, rvc_model, keep_files, yt_url, file_path_input]) generate_btn = gr.Button("Generate", variant='primary') ai_cover = gr.Audio(label='AI Cover') ref_btn.click(update_models_list, None, outputs=rvc_model) inference_mode.change(update_voice_model_visibility, inputs=inference_mode, outputs=rvc_model) is_webui = gr.Number(value=1, visible=False) generate_btn.click(song_cover_pipeline, inputs=[local_file, rvc_model, pitch, keep_files, is_webui, main_gain, backup_gain, inst_gain, index_rate, filter_radius, volume_envelope, f0_method, hop_length, protect, pitch_all, reverb_rm_size, reverb_wet, reverb_dry, reverb_damping, output_format, extra_denoise, inference_mode], outputs=[ai_cover]) 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, True, 'full'], outputs=[pitch, main_gain, backup_gain, inst_gain, index_rate, filter_radius, volume_envelope, protect, f0_method, hop_length, pitch_all, reverb_rm_size, reverb_wet, reverb_dry, reverb_damping, output_format, ai_cover, extra_denoise, inference_mode]) # Download tab with gr.Tab('Download model'): with gr.Tab('From HuggingFace/Pixeldrain URL'): with gr.Row(): model_zip_link = gr.Text(label='Download link to model', info='Should be a zip file containing a .pth model file and an optional .index file.') model_name = gr.Text(label='Name your model', info='Give your new model a unique name from your other voice models.') with gr.Row(): download_btn = gr.Button('Download 🌐', variant='primary', scale=19) dl_output_message = gr.Text(label='Output Message', interactive=False, scale=20) download_btn.click(download_online_model, inputs=[model_zip_link, model_name], outputs=dl_output_message) gr.Markdown('## Input Examples') gr.Examples( [ ['https://huggingface.co/MrDawg/ToothBrushing/resolve/main/ToothBrushing.zip?download=true', 'ToothBrushing'], ['https://huggingface.co/sail-rvc/Aldeano_Minecraft__RVC_V2_-_500_Epochs_/resolve/main/model.pth?download=true, https://huggingface.co/sail-rvc/Aldeano_Minecraft__RVC_V2_-_500_Epochs_/resolve/main/model.index?download=true', 'Minecraft_Villager'], ['https://huggingface.co/phant0m4r/LiSA/resolve/main/LiSA.zip', 'Lisa'], ['https://pixeldrain.com/u/3tJmABXA', 'Gura'], ['https://huggingface.co/Kit-Lemonfoot/kitlemonfoot_rvc_models/resolve/main/AZKi%20(Hybrid).zip', 'Azki'] ], [model_zip_link, model_name], [], download_online_model, cache_examples=False, ) # Upload tab with gr.Tab('Upload model'): gr.Markdown('## Upload locally trained RVC v2 model and index file') gr.Markdown('- Find model file (weights folder) and optional index file (logs/[name] folder)') gr.Markdown('- Compress files into zip file') gr.Markdown('- Upload zip file and give unique name for voice') gr.Markdown('- Click Upload model') with gr.Row(): with gr.Column(): zip_file = gr.File(label='Zip file') local_model_name = gr.Text(label='Model name') with gr.Row(): model_upload_button = gr.Button('Upload model', variant='primary', scale=19) local_upload_output_message = gr.Text(label='Output Message', interactive=False, scale=20) model_upload_button.click(upload_local_model, inputs=[zip_file, local_model_name], outputs=local_upload_output_message) app.launch( share=args.share_enabled, debug=args.share_enabled, show_error=True, server_name=None if not args.listen else (args.listen_host or '0.0.0.0'), server_port=args.listen_port, ssr_mode=args.ssr )