ort nightly

app.py

@@ -1,287 +1,4 @@
-import json
 import os
-import shutil
-import urllib.request
-import zipfile
-import gdown
-from argparse import ArgumentParser
 
-import gradio as gr
-import spaces
-
-from src.main import song_cover_pipeline
-
-BASE_DIR = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
-
-mdxnet_models_dir = 'mdxnet_models'
-rvc_models_dir = 'rvc_models'
-output_dir = 'song_output'
-
-def download_and_extract_model(model_url, model_name, progress=gr.Progress()):
-    try:
-        os.makedirs(rvc_models_dir, exist_ok=True)
-
-        extraction_folder = os.path.join(rvc_models_dir, model_name)
-        zip_path = os.path.join(rvc_models_dir, f'{model_name}.zip')
-        
-        if os.path.exists(extraction_folder):
-            raise gr.Error(f'Voice model directory {model_name} already exists! Choose a different name for your voice model.')
-        
-        progress(0, desc=f'[~] Downloading voice model with name {model_name}...')
-        
-        try:
-            if 'huggingface.co' in model_url:
-                urllib.request.urlretrieve(model_url, zip_path)
-            elif 'pixeldrain.com' in model_url:
-                pixeldrain_id = model_url.split('/')[-1]
-                pixeldrain_url = f'https://pixeldrain.com/api/file/{pixeldrain_id}'
-                urllib.request.urlretrieve(pixeldrain_url, zip_path)
-            elif 'drive.google.com' in model_url:
-                file_id = model_url.split('/')[-2]
-                gdown.download(id=file_id, output=zip_path, quiet=False)
-            else:
-                urllib.request.urlretrieve(model_url, zip_path)
-        except Exception as download_error:
-            raise gr.Error(f"Failed to download the model: {str(download_error)}")
-
-        if not os.path.exists(zip_path):
-            raise gr.Error(f"Failed to download the model. The zip file was not created.")
-
-        progress(0.5, desc="Extracting model...")
-        extract_zip(extraction_folder, zip_path)
-        
-        pth_files = [f for f in os.listdir(extraction_folder) if f.endswith('.pth')]
-        if not pth_files:
-            raise ValueError("No .pth file found in the downloaded model.")
-        
-        progress(1, desc="Model ready")
-        return model_name
-
-    except Exception as e:
-        if os.path.exists(extraction_folder):
-            shutil.rmtree(extraction_folder)
-        if os.path.exists(zip_path):
-            os.remove(zip_path)
-        raise gr.Error(f"Error downloading or extracting model: {str(e)}")
-
-def cleanup_temp_model(model_name):
-    temp_dir = os.path.join(rvc_models_dir, model_name)
-    try:
-        shutil.rmtree(temp_dir)
-    except Exception as e:
-        print(f"Error cleaning up temporary model files: {str(e)}")
-
-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 'huggingface.co' in url:
-            urllib.request.urlretrieve(url, zip_name)
-
-        if 'pixeldrain.com' in url:
-            zip_name = dir_name + '.zip'
-            url = f'https://pixeldrain.com/api/file/{zip_name}'
-            urllib.request.urlretrieve(url, zip_name)
-
-        elif 'drive.google.com' in url:
-            # Extract the Google Drive file ID
-            zip_name = dir_name + '.zip'
-            file_id = url.split('/')[-2]
-            output = os.path.join('.', f'{dir_name}.zip')  # Adjust the output path if needed
-            gdown.download(id=file_id, output=output, quiet=False)
-
-        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 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)
-
-
-def song_cover_pipeline_with_model_download(song_input, model_url, model_name, 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, progress=gr.Progress()):
-    model_path = None
-    try:
-        model_path = download_and_extract_model(model_url, model_name, progress)
-        print(f"Model path: {model_path}")
-        result = song_cover_pipeline(song_input, model_path, 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, progress)
-        
-        # Clean up old folders in song_output
-        output_folders = [f for f in os.listdir(output_dir) if os.path.isdir(os.path.join(output_dir, f))]
-        output_folders.sort(key=lambda x: os.path.getmtime(os.path.join(output_dir, x)))
-        
-        while len(output_folders) > 100:
-            oldest_folder = output_folders.pop(0)
-            shutil.rmtree(os.path.join(output_dir, oldest_folder))
-        
-        return result
-    except gr.Error as e:
-        return str(e), None  # Return error message and None for the second output
-    finally:
-        if model_path:
-            cleanup_temp_model(model_path)
-
-
-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()
-
-    with gr.Blocks(title='AICoverGenWebUI', theme='NoCrypt/miku@1.2.2') as app:
-
-        gr.Label('AICoverGen WebUI created with ❤️', show_label=False)
-
-        # main tab
-        with gr.Tab("Generate"):
-
-            with gr.Accordion('Main Options'):
-                with gr.Row():
-                    with gr.Column():
-                        model_url = gr.Text(label='Voice Model URL', info='Enter the URL of the voice model zip file', value='https://huggingface.co/megaaziib/my-rvc-models-collection/resolve/main/kobo.zip')
-                        model_name = gr.Text(label='Voice Model Name', info='Enter the name of the voice model', value='kobo')
-                        # 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')
-
-                    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.', value='https://youtu.be/FRh7LvlQTuA')
-                        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(-24, 24, value=0, step=1, label='Pitch Change (Vocals ONLY)', info='Generally, use 12 for male to female conversions and -12 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, model_url, keep_files, local_file])
-                generate_btn = gr.Button("Generate", variant='primary')
-            with gr.Row():
-                ai_cover = gr.Audio(label='AI Cover (Vocal Only Inference)', show_share_button=False)
-                ai_backing = gr.Audio(label='AI Cover (Vocal Backing Inference)', show_share_button=False)
-
-            is_webui = gr.Number(value=1, visible=False)
-            generate_btn.click(song_cover_pipeline_with_model_download,
-                inputs=[song_input, model_url, model_name, 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, ai_backing])
-            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])
-
-    app.launch(
-        share=args.share_enabled,
-        server_name=None if not args.listen else (args.listen_host or '0.0.0.0'),
-        server_port=args.listen_port,
-    )
+os.system("pip install ort-nightly-gpu --index-url=https://aiinfra.pkgs.visualstudio.com/PublicPackages/_packaging/ort-cuda-12-nightly/pypi/simple/")
+os.system("python src/webui.py")

requirements.txt

@@ -20,4 +20,5 @@ tqdm==4.65.0
 yt_dlp==2024.8.6
 sox==1.4.1
 audio-separator[gpu]==0.17.5
-gdown==5.2.0
+gdown==5.2.0
+ort-nightly-gpu @ https://aiinfra.pkgs.visualstudio.com/PublicPackages/_packaging/ort-cuda-12-nightly/pypi/simple/ort-nightly-gpu/

src/infer_pack/attentions.py

@@ -5,9 +5,9 @@ import torch
 from torch import nn
 from torch.nn import functional as F
 
-from src.infer_pack import commons
-from src.infer_pack import modules
-from src.infer_pack.modules import LayerNorm
+from infer_pack import commons
+from infer_pack import modules
+from infer_pack.modules import LayerNorm
 
 
 class Encoder(nn.Module):

src/infer_pack/models.py

@@ -3,15 +3,15 @@ from time import time as ttime
 import torch
 from torch import nn
 from torch.nn import functional as F
-from src.infer_pack import modules
-from src.infer_pack import attentions
-from src.infer_pack import commons
-from src.infer_pack.commons import init_weights, get_padding
+from infer_pack import modules
+from infer_pack import attentions
+from infer_pack import commons
+from infer_pack.commons import init_weights, get_padding
 from torch.nn import Conv1d, ConvTranspose1d, AvgPool1d, Conv2d
 from torch.nn.utils import weight_norm, remove_weight_norm, spectral_norm
-from src.infer_pack.commons import init_weights
+from infer_pack.commons import init_weights
 import numpy as np
-from src.infer_pack import commons
+from infer_pack import commons
 
 
 class TextEncoder256(nn.Module):

src/infer_pack/modules.py

@@ -9,9 +9,9 @@ from torch.nn import functional as F
 from torch.nn import Conv1d, ConvTranspose1d, AvgPool1d, Conv2d
 from torch.nn.utils import weight_norm, remove_weight_norm
 
-from src.infer_pack import commons
-from src.infer_pack.commons import init_weights, get_padding
-from src.infer_pack.transforms import piecewise_rational_quadratic_transform
+from infer_pack import commons
+from infer_pack.commons import init_weights, get_padding
+from infer_pack.transforms import piecewise_rational_quadratic_transform
 
 
 LRELU_SLOPE = 0.1

src/main.py

@@ -18,8 +18,8 @@ from pedalboard import Pedalboard, Reverb, Compressor, HighpassFilter
 from pedalboard.io import AudioFile
 from pydub import AudioSegment
 
-from src.mdx import run_mdx, run_roformer
-from src.rvc import Config, load_hubert, get_vc, rvc_infer
+from mdx import run_mdx, run_roformer
+from rvc import Config, load_hubert, get_vc, rvc_infer
 
 import spaces
 

src/rvc.py

@@ -4,14 +4,14 @@ from pathlib import Path
 import torch
 from scipy.io import wavfile
 
-from src.infer_pack.models import (
+from infer_pack.models import (
     SynthesizerTrnMs256NSFsid,
     SynthesizerTrnMs256NSFsid_nono,
     SynthesizerTrnMs768NSFsid,
     SynthesizerTrnMs768NSFsid_nono,
 )
-from src.my_utils import load_audio
-from src.vc_infer_pipeline import VC
+from my_utils import load_audio
+from vc_infer_pipeline import VC
 
 BASE_DIR = Path(__file__).resolve().parent.parent
 

src/webui.py

@@ -0,0 +1,287 @@
+import json
+import os
+import shutil
+import urllib.request
+import zipfile
+import gdown
+from argparse import ArgumentParser
+
+import gradio as gr
+import spaces
+
+from main import song_cover_pipeline
+
+BASE_DIR = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
+
+mdxnet_models_dir = 'mdxnet_models'
+rvc_models_dir = 'rvc_models'
+output_dir = 'song_output'
+
+def download_and_extract_model(model_url, model_name, progress=gr.Progress()):
+    try:
+        os.makedirs(rvc_models_dir, exist_ok=True)
+
+        extraction_folder = os.path.join(rvc_models_dir, model_name)
+        zip_path = os.path.join(rvc_models_dir, f'{model_name}.zip')
+        
+        if os.path.exists(extraction_folder):
+            raise gr.Error(f'Voice model directory {model_name} already exists! Choose a different name for your voice model.')
+        
+        progress(0, desc=f'[~] Downloading voice model with name {model_name}...')
+        
+        try:
+            if 'huggingface.co' in model_url:
+                urllib.request.urlretrieve(model_url, zip_path)
+            elif 'pixeldrain.com' in model_url:
+                pixeldrain_id = model_url.split('/')[-1]
+                pixeldrain_url = f'https://pixeldrain.com/api/file/{pixeldrain_id}'
+                urllib.request.urlretrieve(pixeldrain_url, zip_path)
+            elif 'drive.google.com' in model_url:
+                file_id = model_url.split('/')[-2]
+                gdown.download(id=file_id, output=zip_path, quiet=False)
+            else:
+                urllib.request.urlretrieve(model_url, zip_path)
+        except Exception as download_error:
+            raise gr.Error(f"Failed to download the model: {str(download_error)}")
+
+        if not os.path.exists(zip_path):
+            raise gr.Error(f"Failed to download the model. The zip file was not created.")
+
+        progress(0.5, desc="Extracting model...")
+        extract_zip(extraction_folder, zip_path)
+        
+        pth_files = [f for f in os.listdir(extraction_folder) if f.endswith('.pth')]
+        if not pth_files:
+            raise ValueError("No .pth file found in the downloaded model.")
+        
+        progress(1, desc="Model ready")
+        return model_name
+
+    except Exception as e:
+        if os.path.exists(extraction_folder):
+            shutil.rmtree(extraction_folder)
+        if os.path.exists(zip_path):
+            os.remove(zip_path)
+        raise gr.Error(f"Error downloading or extracting model: {str(e)}")
+
+def cleanup_temp_model(model_name):
+    temp_dir = os.path.join(rvc_models_dir, model_name)
+    try:
+        shutil.rmtree(temp_dir)
+    except Exception as e:
+        print(f"Error cleaning up temporary model files: {str(e)}")
+
+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 'huggingface.co' in url:
+            urllib.request.urlretrieve(url, zip_name)
+
+        if 'pixeldrain.com' in url:
+            zip_name = dir_name + '.zip'
+            url = f'https://pixeldrain.com/api/file/{zip_name}'
+            urllib.request.urlretrieve(url, zip_name)
+
+        elif 'drive.google.com' in url:
+            # Extract the Google Drive file ID
+            zip_name = dir_name + '.zip'
+            file_id = url.split('/')[-2]
+            output = os.path.join('.', f'{dir_name}.zip')  # Adjust the output path if needed
+            gdown.download(id=file_id, output=output, quiet=False)
+
+        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 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)
+
+
+def song_cover_pipeline_with_model_download(song_input, model_url, model_name, 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, progress=gr.Progress()):
+    model_path = None
+    try:
+        model_path = download_and_extract_model(model_url, model_name, progress)
+        print(f"Model path: {model_path}")
+        result = song_cover_pipeline(song_input, model_path, 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, progress)
+        
+        # Clean up old folders in song_output
+        output_folders = [f for f in os.listdir(output_dir) if os.path.isdir(os.path.join(output_dir, f))]
+        output_folders.sort(key=lambda x: os.path.getmtime(os.path.join(output_dir, x)))
+        
+        while len(output_folders) > 100:
+            oldest_folder = output_folders.pop(0)
+            shutil.rmtree(os.path.join(output_dir, oldest_folder))
+        
+        return result
+    except gr.Error as e:
+        return str(e), None  # Return error message and None for the second output
+    finally:
+        if model_path:
+            cleanup_temp_model(model_path)
+
+
+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()
+
+    with gr.Blocks(title='AICoverGenWebUI', theme='NoCrypt/miku@1.2.2') as app:
+
+        gr.Label('AICoverGen WebUI created with ❤️', show_label=False)
+
+        # main tab
+        with gr.Tab("Generate"):
+
+            with gr.Accordion('Main Options'):
+                with gr.Row():
+                    with gr.Column():
+                        model_url = gr.Text(label='Voice Model URL', info='Enter the URL of the voice model zip file', value='https://huggingface.co/megaaziib/my-rvc-models-collection/resolve/main/kobo.zip')
+                        model_name = gr.Text(label='Voice Model Name', info='Enter the name of the voice model', value='kobo')
+                        # 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')
+
+                    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.', value='https://youtu.be/FRh7LvlQTuA')
+                        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(-24, 24, value=0, step=1, label='Pitch Change (Vocals ONLY)', info='Generally, use 12 for male to female conversions and -12 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, model_url, keep_files, local_file])
+                generate_btn = gr.Button("Generate", variant='primary')
+            with gr.Row():
+                ai_cover = gr.Audio(label='AI Cover (Vocal Only Inference)', show_share_button=False)
+                ai_backing = gr.Audio(label='AI Cover (Vocal Backing Inference)', show_share_button=False)
+
+            is_webui = gr.Number(value=1, visible=False)
+            generate_btn.click(song_cover_pipeline_with_model_download,
+                inputs=[song_input, model_url, model_name, 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, ai_backing])
+            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])
+
+    app.launch(
+        share=args.share_enabled,
+        server_name=None if not args.listen else (args.listen_host or '0.0.0.0'),
+        server_port=args.listen_port,
+    )