File size: 18,037 Bytes
96ea114
 
 
 
 
 
6c4de84
96ea114
 
 
 
6c4de84
400af59
96ea114
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
e530855
96ea114
e530855
96ea114
d6f6f3b
824b940
3e80b7b
96ea114
 
 
 
 
 
 
 
 
23d230d
e530855
 
 
 
32dc285
 
 
 
96ea114
 
 
 
 
 
 
 
 
424f7a9
96ea114
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
cd8f7a7
96ea114
 
 
 
 
32dc285
96ea114
 
 
 
424f7a9
96ea114
 
 
400af59
4cc2e85
400af59
 
1d971a6
400af59
0726a41
 
96ea114
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
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
from my_utils import show_stored_files
from my_utils import remove_files_and_folders

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 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='ReVox') as app:

        gr.Label('ReVox WebUI created with much much ❤️', show_label=False)

        gr.Markdown("Duplicate the space for use in private")
        gr.Markdown("[![Duplicate this Space](https://huggingface.co/datasets/huggingface/badges/raw/main/duplicate-this-space-sm-dark.svg)](https://huggingface.co/spaces/r3gm/AICoverGen?duplicate=true)\n\n") 
        
        # main 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 file_upload_col:
                        main_vocals_input = gr.Text(label='main_vocals_input')
                        backup_vocals_input = gr.Text(label='backup_vocals_input')
                        main_vocals = gr.File(label='Audio file')
                        backup_vocals = gr.File(label='Backup Vocals')
                        main_vocals_file = gr.UploadButton('Upload Main Dereverbered vocals', file_types=['audio'], variant='primary')
                        main_vocals_file.upload(process_file_upload, inputs=[main_vocals_file], outputs=[main_vocals, main_vocals_input])
                        backup_vocals_file = gr.UploadButton('Upload backup vocals', file_types=['audio'], variant='primary')
                        backup_vocals_file.upload(process_file_upload, inputs=[backup_vocals_file], outputs=[backup_vocals, backup_vocals_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.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, 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=[main_vocals_input, backup_vocals_input, rvc_model, keep_files, main_vocals, backup_vocals])
                generate_btn = gr.Button("Generate", 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=[main_vocals_input, backup_vocals_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, 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])
            
        
            with gr.Row():
                clear_cache_button = gr.Button(value='Clear Gradio Cache')
            clear_cache_button.click(remove_files_and_folders)
            
            


        # 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/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,
                )

            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,
    )