import json import os import shutil import urllib.request import zipfile from argparse import ArgumentParser import gradio as gr from main import song_cover_pipeline BASE_DIR = os.path.dirname(os.path.dirname(os.path.abspath(__file__))) mdxnet_models_dir = os.path.join(BASE_DIR, 'mdxnet_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 = ['hubert_base.pt', 'MODELS.txt', 'public_models.json', 'rmvpe.pt'] 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.Dropdown.update(choices=models_l) def load_public_models(): models_table = [] for model in public_models['voice_models']: if not model['name'] in voice_models: model = [model['name'], model['description'], model['credit'], model['url'], ', '.join(model['tags'])] models_table.append(model) tags = list(public_models['tags'].keys()) return gr.DataFrame.update(value=models_table), gr.CheckboxGroup.update(choices=tags) 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}.') # move model and index file to 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))) # remove any unnecessary nested folders 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}' 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.') zip_name = 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_link = "https://youtu.be/hT_nvWreIhg" def filter_models(tags, query): models_table = [] # no filter if len(tags) == 0 and len(query) == 0: for model in public_models['voice_models']: models_table.append([model['name'], model['description'], model['credit'], model['url'], model['tags']]) # filter based on tags and query elif len(tags) > 0 and len(query) > 0: for model in public_models['voice_models']: if all(tag in model['tags'] for tag in tags): model_attributes = f"{model['name']} {model['description']} {model['credit']} {' '.join(model['tags'])}".lower() if query.lower() in model_attributes: models_table.append([model['name'], model['description'], model['credit'], model['url'], model['tags']]) # filter based on only tags elif len(tags) > 0: for model in public_models['voice_models']: if all(tag in model['tags'] for tag in tags): models_table.append([model['name'], model['description'], model['credit'], model['url'], model['tags']]) # filter based on only query else: for model in public_models['voice_models']: model_attributes = f"{model['name']} {model['description']} {model['credit']} {' '.join(model['tags'])}".lower() if query.lower() in model_attributes: models_table.append([model['name'], model['description'], model['credit'], model['url'], model['tags']]) return gr.DataFrame.update(value=models_table) def pub_dl_autofill(pub_models, event: gr.SelectData): return gr.Text.update(value=pub_models.loc[event.index[0], 'URL']), gr.Text.update(value=pub_models.loc[event.index[0], 'Model Name']) def swap_visibility(): return gr.update(visible=True), gr.update(visible=False), gr.update(value=''), gr.update(value=None) def process_file_upload(file): return file.name, gr.update(value=file.name) def show_hop_slider(pitch_detection_algo): if pitch_detection_algo == 'mangio-crepe': return gr.update(visible=True) else: return gr.update(visible=False) 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.') args = parser.parse_args() voice_models = get_current_models(rvc_models_dir) with open(os.path.join(rvc_models_dir, 'public_models.json'), encoding='utf8') as infile: public_models = json.load(infile) with gr.Blocks(title='oItsMineZ\'s AI Cover WebUI', theme=gr.themes.Base(font=[gr.themes.GoogleFont("Noto Sans Thai"), "sans-serif"])) as app: gr.Label('oItsMineZ\'s RVC v2 AI Cover WebUI', show_label=False) gr.Markdown( "
\n\n"+ "RVC v2 Model"+ "[![oItsMineZ's RVC Model](https://img.shields.io/badge/%F0%9F%A4%97_Hugging_Face-_oItsMineZ's%20RVC%20%20Model-yellow?style=for-the-badge&logoColor=yellow)](https://huggingface.co/oItsMineZ/oItsMineZ-RVC-Model)\n\n"+ "ติดตาม oItsMineZ"+ "[![oItsMineZ on YouTube](https://img.shields.io/badge/YouTube-FF0000?style=for-the-badge&logo=youtube&logoColor=white)](https://www.youtube.com/@oItsMineZ?sub_confirmation=1)"+ "
" ) # Main Tab with gr.Tab("📢 Info"): gr.Markdown("## 📌แนะนำให้โคลน Space นี้ไว้ในบัญชีของคุณ เพื่อการใช้งานที่ดียิ่งขึ้น (ต้องสมัครบัญชี Hugging Face ก่อน)") gr.Markdown("[![Duplicate this Space](https://huggingface.co/datasets/huggingface/badges/raw/main/duplicate-this-space-sm-dark.svg)](https://huggingface.co/spaces/oItsMineZ/RVC-v2-AI-Cover-WebUI?duplicate=true)\n\n") gr.HTML("

📄ข้อควรรู้

") gr.Markdown("- RVC v2 (Retrieval Based Voice Conversion v2) เป็น AI Voice Model ที่ปรับปรุงมาจาก VITS ที่ทำให้เทรนโมเดลได้ง่ายขึ้น และคุณภาพของเสียงดีขึ้น") gr.Markdown("- WebUI นี้ใช้สำหรับเฉพาะ **AI Cover** เพลงเท่านั้น! ถ้าอยากใช้เฉพาะเสียงพูดให้ใช้ [**ตัวนี้แทน**](https://huggingface.co/spaces/oItsMineZ/RVC-v2-WebUI)") gr.Markdown("- บางเพลงอาจ**ใช้เวลานานมากๆ** ขึ้นอยู่กับความยาวหรือขนาดไฟล์เพลง (จากที่ผมลองเพลงของ [**OneRepublic - Counting Stars**](https://youtu.be/hT_nvWreIhg) ใช้เวลา 55 นาที 😱)") gr.Markdown("- ถ้าใช้ในโทรศัพท์ **ห้าม**ออกจากหน้า Web ขณะเว็บกำลังดำเนินการอยู่ เพราะทำให้ไฟล์หายระหว่างขั้นตอนได้") gr.Markdown("- ถ้าคุณพร้อมที่จะทำเพลง AI Cover แล้ว ให้คลิกแท็บ 🎵 Generate ได้เลย!") gr.HTML("

✨ฟีเจอร์

") gr.Markdown("- นำเพลงจาก YouTube มา Cover ได้ทันที เพียงแค่ก็อบลิงก์มาวาง") gr.Markdown("- อัปโหลดไฟล์เพลงของตัวเองได้เลย โดยไม่ต้องลบทำนองออกก่อน") gr.Markdown("- มี UVR5 (Ultimate Vocal Remover v5) โดยช่วยแยกเสียงร้องกับทำนองออกจากเพลง") gr.Markdown("- สามารถดาวน์โหลด Model อื่นๆ ได้ที่แท็บ ⬇️ Download Model [**(เว็บสำหรับหา Model เพิ่มเติม)**](https://voice-models.com)") gr.Markdown("- ที่สำคัญ **อย่าลืม** *Refresh Model* ทุกครั้งเมื่อโหลด Model ใหม่เข้ามา") gr.HTML("

📋รายชื่อ Model

") gr.Markdown("- อาจารย์แดง (DaengGuitar) - 500 Epochs") gr.Markdown("- เต้ (TAEEXZENFIRE) - 500 Epochs") gr.Markdown("- ท่านศาสดา - 50 Epochs") gr.Markdown("- Model ใหม่เร็วๆ นี้ 🤫") gr.HTML("

🌐WebUI อื่นๆ

") gr.Markdown("- เฉพาะ Vocal (เสียงพูดปกติ)") gr.Markdown("[![Hugging Face Spaces](https://img.shields.io/badge/%F0%9F%8E%99%EF%B8%8F_Space-_oItsMineZ's%20RVC%20v2%20WebUI-orange?style=for-the-badge)](https://huggingface.co/spaces/oItsMineZ/RVC-v2-WebUI)") gr.HTML("

📱เวอร์ชั่นอื่นๆ

") gr.Markdown("- Google Colab (No WebUI)") gr.Markdown("[![Open in Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/oItsMineZ/RVC-v2-AICover-Colab/blob/main/oItsMineZ-rvc-v2-AICover-Colab.ipynb)") gr.HTML("

❤️ขอขอบคุณ

") gr.Markdown("- [**@SociallyIneptWeeb**](https://github.com/SociallyIneptWeeb) for [***AICoverGen***](https://github.com/SociallyIneptWeeb/AICoverGen)") # Generate Tab with gr.Tab("🎵 Generate"): with gr.Accordion('Main Options'): 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 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. Example: https://youtu.be/dQw4w9WgXcQ', value=def_link) show_file_upload_button = gr.Button('Upload file instead 📤') with gr.Column(visible=False) as file_upload_col: local_file = gr.File(label='Audio file') song_input_file = gr.UploadButton('Upload 📂', file_types=['audio'], variant='primary') show_yt_link_button = gr.Button('Paste YouTube link/Path to local file instead 🔗') 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)') show_file_upload_button.click(swap_visibility, outputs=[file_upload_col, yt_link_col, song_input, local_file]) show_yt_link_button.click(swap_visibility, outputs=[yt_link_col, file_upload_col, song_input, local_file]) with gr.Accordion('Voice conversion options', open=False): with gr.Row(): 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 median filtering to the harvested pitch results. Can reduce breathiness') rms_mix_rate = gr.Slider(0, 1, value=0.25, label='RMS mix rate', 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', 'mangio-crepe'], value='rmvpe', label='Pitch detection algorithm', info='Best option is rmvpe (clarity in vocals), then mangio-crepe (smoother vocals)') crepe_hop_length = gr.Slider(32, 320, value=128, step=1, visible=False, label='Crepe 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=crepe_hop_length) keep_files = gr.Checkbox(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') 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(): clear_btn = gr.ClearButton(value='Clear', components=[song_input, rvc_model, keep_files, local_file]) generate_btn = gr.Button("✨ Generate Song ✨", variant='primary') ai_cover = gr.Audio(label='AI Cover', 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(song_cover_pipeline, inputs=[song_input, rvc_model, pitch, keep_files, is_webui, main_gain, backup_gain, inst_gain, index_rate, filter_radius, rms_mix_rate, f0_method, crepe_hop_length, protect, pitch_all, reverb_rm_size, reverb_wet, reverb_dry, reverb_damping, output_format], 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], 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'): with gr.Tab('From HuggingFace/Pixeldrain URL'): with gr.Row(): model_name = gr.Text(label='Model Name', info='Give your new model a unique name from your other voice models.') model_zip_link = gr.Text(label='Link', info='Should be a zip file containing a .pth model file and an optional .index file.') 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('## oItsMineZ\'s Model List') gr.Examples( [ ['DaengGuitar', 'https://huggingface.co/oItsMineZ/oItsMineZ-RVC-Model/resolve/main/DaengGuitar/DaengGuitar.zip'], ['TAEEXZENFIRE', 'https://huggingface.co/oItsMineZ/oItsMineZ-RVC-Model/resolve/main/TAEEXZENFIRE/TAEEXZENFIRE.zip'], ['ท่านศาสดา', 'https://huggingface.co/oItsMineZ/oItsMineZ-RVC-Model/resolve/main/Sadsada/Sadsada.zip'] ], [model_name, model_zip_link], [], download_online_model, ) with gr.Tab('From Public Index'): gr.Markdown('## How to use') gr.Markdown('- Click Initialize public models table') gr.Markdown('- Filter models using tags or search bar') gr.Markdown('- Select a row to autofill the download link and model name') gr.Markdown('- Click Download') with gr.Row(): pub_zip_link = gr.Text(label='Download link to model') pub_model_name = gr.Text(label='Model name') with gr.Row(): download_pub_btn = gr.Button('Download 🌐', variant='primary', scale=19) pub_dl_output_message = gr.Text(label='Output Message', interactive=False, scale=20) filter_tags = gr.CheckboxGroup(value=[], label='Show voice models with tags', choices=[]) search_query = gr.Text(label='Search') load_public_models_button = gr.Button(value='Initialize public models table', variant='primary') public_models_table = gr.DataFrame(value=[], headers=['Model Name', 'Description', 'Credit', 'URL', 'Tags'], label='Available Public Models', interactive=False) public_models_table.select(pub_dl_autofill, inputs=[public_models_table], outputs=[pub_zip_link, pub_model_name]) load_public_models_button.click(load_public_models, outputs=[public_models_table, filter_tags]) search_query.change(filter_models, inputs=[filter_tags, search_query], outputs=public_models_table) filter_tags.change(filter_models, inputs=[filter_tags, search_query], outputs=public_models_table) download_pub_btn.click(download_online_model, inputs=[pub_zip_link, pub_model_name], outputs=pub_dl_output_message) # 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, enable_queue=True, server_name=None if not args.listen else (args.listen_host or '0.0.0.0'), server_port=args.listen_port, )