helloWorld199 commited on
Commit
13f9cfc
1 Parent(s): 6172477

Upload main.py

Browse files
Files changed (1) hide show
  1. main.py +438 -0
main.py ADDED
@@ -0,0 +1,438 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import argparse
2
+ import gc
3
+ import hashlib
4
+ import json
5
+ import os
6
+ import shlex
7
+ import subprocess
8
+ from contextlib import suppress
9
+ from urllib.parse import urlparse, parse_qs
10
+ import time
11
+
12
+ import gradio as gr
13
+ import librosa
14
+ import numpy as np
15
+ import soundfile as sf
16
+ import sox
17
+ import yt_dlp
18
+ from pedalboard import Pedalboard, Reverb, Compressor, HighpassFilter
19
+ from pedalboard.io import AudioFile
20
+ from pydub import AudioSegment
21
+ from my_utils import add_stem_name
22
+
23
+ from mdx import run_mdx
24
+ from rvc import Config, load_hubert, get_vc, rvc_infer
25
+
26
+ BASE_DIR = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
27
+
28
+ mdxnet_models_dir = os.path.join(BASE_DIR, 'mdxnet_models')
29
+ rvc_models_dir = os.path.join(BASE_DIR, 'rvc_models')
30
+ output_dir = os.path.join(BASE_DIR, 'song_output')
31
+
32
+
33
+ def get_youtube_video_id(url, ignore_playlist=True):
34
+ """
35
+ Examples:
36
+ http://youtu.be/SA2iWivDJiE
37
+ http://www.youtube.com/watch?v=_oPAwA_Udwc&feature=feedu
38
+ http://www.youtube.com/embed/SA2iWivDJiE
39
+ http://www.youtube.com/v/SA2iWivDJiE?version=3&hl=en_US
40
+ """
41
+ query = urlparse(url)
42
+ if query.hostname == 'youtu.be':
43
+ if query.path[1:] == 'watch':
44
+ return query.query[2:]
45
+ return query.path[1:]
46
+
47
+ if query.hostname in {'www.youtube.com', 'youtube.com', 'music.youtube.com'}:
48
+ if not ignore_playlist:
49
+ # use case: get playlist id not current video in playlist
50
+ with suppress(KeyError):
51
+ return parse_qs(query.query)['list'][0]
52
+ if query.path == '/watch':
53
+ return parse_qs(query.query)['v'][0]
54
+ if query.path[:7] == '/watch/':
55
+ return query.path.split('/')[1]
56
+ if query.path[:7] == '/embed/':
57
+ return query.path.split('/')[2]
58
+ if query.path[:3] == '/v/':
59
+ return query.path.split('/')[2]
60
+
61
+ # returns None for invalid YouTube url
62
+ return None
63
+
64
+
65
+ def yt_download(link):
66
+ ydl_opts = {
67
+ 'format': 'bestaudio',
68
+ 'outtmpl': '%(title)s',
69
+ 'nocheckcertificate': True,
70
+ 'ignoreerrors': True,
71
+ 'no_warnings': True,
72
+ 'quiet': True,
73
+ 'extractaudio': True,
74
+ 'postprocessors': [{'key': 'FFmpegExtractAudio', 'preferredcodec': 'mp3'}],
75
+ }
76
+ with yt_dlp.YoutubeDL(ydl_opts) as ydl:
77
+ result = ydl.extract_info(link, download=True)
78
+ download_path = ydl.prepare_filename(result, outtmpl='%(title)s.mp3')
79
+
80
+ return download_path
81
+
82
+
83
+ def raise_exception(error_msg, is_webui):
84
+ if is_webui:
85
+ raise gr.Error(error_msg)
86
+ else:
87
+ raise Exception(error_msg)
88
+
89
+
90
+ def get_rvc_model(voice_model, is_webui):
91
+ rvc_model_filename, rvc_index_filename = None, None
92
+ model_dir = os.path.join(rvc_models_dir, voice_model)
93
+ for file in os.listdir(model_dir):
94
+ ext = os.path.splitext(file)[1]
95
+ if ext == '.pth':
96
+ rvc_model_filename = file
97
+ if ext == '.index':
98
+ rvc_index_filename = file
99
+
100
+ if rvc_model_filename is None:
101
+ error_msg = f'No model file exists in {model_dir}.'
102
+ raise_exception(error_msg, is_webui)
103
+
104
+ return os.path.join(model_dir, rvc_model_filename), os.path.join(model_dir, rvc_index_filename) if rvc_index_filename else ''
105
+
106
+
107
+ def get_audio_paths(song_dir):
108
+ orig_song_path = None
109
+ instrumentals_path = None
110
+ main_vocals_dereverb_path = None
111
+ backup_vocals_path = None
112
+
113
+ for file in os.listdir(song_dir):
114
+ if file.endswith('_Instrumental.wav'):
115
+ instrumentals_path = os.path.join(song_dir, file)
116
+ orig_song_path = instrumentals_path.replace('_Instrumental', '')
117
+
118
+ elif file.endswith('_Vocals_Main_DeReverb.wav'):
119
+ main_vocals_dereverb_path = os.path.join(song_dir, file)
120
+
121
+ elif file.endswith('_Vocals_Backup.wav'):
122
+ backup_vocals_path = os.path.join(song_dir, file)
123
+
124
+ return orig_song_path, instrumentals_path, main_vocals_dereverb_path, backup_vocals_path
125
+
126
+
127
+ def convert_to_stereo(audio_path):
128
+ wave, sr = librosa.load(audio_path, mono=False, sr=44100)
129
+
130
+ # check if mono
131
+ if type(wave[0]) != np.ndarray:
132
+ stereo_path = f'{os.path.splitext(audio_path)[0]}_stereo.wav'
133
+ command = shlex.split(f'ffmpeg -y -loglevel error -i "{audio_path}" -ac 2 -f wav "{stereo_path}"')
134
+ subprocess.run(command)
135
+ return stereo_path
136
+ else:
137
+ return audio_path
138
+
139
+
140
+ def pitch_shift(audio_path, pitch_change):
141
+ output_path = f'{os.path.splitext(audio_path)[0]}_p{pitch_change}.wav'
142
+ if not os.path.exists(output_path):
143
+ y, sr = sf.read(audio_path)
144
+ tfm = sox.Transformer()
145
+ tfm.pitch(pitch_change)
146
+ y_shifted = tfm.build_array(input_array=y, sample_rate_in=sr)
147
+ sf.write(output_path, y_shifted, sr)
148
+
149
+ return output_path
150
+
151
+
152
+ def get_hash(filepath):
153
+ with open(filepath, 'rb') as f:
154
+ file_hash = hashlib.blake2b()
155
+ while chunk := f.read(8192):
156
+ file_hash.update(chunk)
157
+
158
+ return file_hash.hexdigest()[:11]
159
+
160
+
161
+ def display_progress(message, percent, is_webui, progress=None):
162
+ if is_webui:
163
+ progress(percent, desc=message)
164
+ else:
165
+ print(message)
166
+
167
+
168
+ def preprocess_song(song_input, mdx_model_params, song_id, is_webui, input_type, progress=None):
169
+ keep_orig = False
170
+ if input_type == 'yt':
171
+ display_progress('[~] Downloading song...', 0, is_webui, progress)
172
+ song_link = song_input.split('&')[0]
173
+ orig_song_path = yt_download(song_link)
174
+ elif input_type == 'local':
175
+ orig_song_path = song_input
176
+ keep_orig = True
177
+ else:
178
+ orig_song_path = None
179
+
180
+ song_output_dir = os.path.join(output_dir, song_id)
181
+ orig_song_path = convert_to_stereo(orig_song_path)
182
+
183
+ display_progress('[~] Separating Vocals from Instrumental...', 0.1, is_webui, progress)
184
+ vocals_path, instrumentals_path = run_mdx(mdx_model_params, song_output_dir, os.path.join(mdxnet_models_dir, 'UVR-MDX-NET-Voc_FT.onnx'), orig_song_path, denoise=True, keep_orig=keep_orig, _stemname1 = "_origvocals", _stemname2="_originstr")
185
+
186
+ display_progress('[~] Separating Main Vocals from Backup Vocals...', 0.2, is_webui, progress)
187
+ backup_vocals_path, main_vocals_path = run_mdx(mdx_model_params, song_output_dir, os.path.join(mdxnet_models_dir, 'UVR_MDXNET_KARA_2.onnx'), vocals_path, suffix='Backup', invert_suffix='Main', denoise=True)
188
+
189
+ display_progress('[~] Applying DeReverb to Vocals...', 0.3, is_webui, progress)
190
+ _, main_vocals_dereverb_path = run_mdx(mdx_model_params, song_output_dir, os.path.join(mdxnet_models_dir, 'Reverb_HQ_By_FoxJoy.onnx'), main_vocals_path, invert_suffix='DeReverb', exclude_main=True, denoise=True)
191
+
192
+ return orig_song_path, vocals_path, instrumentals_path, main_vocals_path, backup_vocals_path, main_vocals_dereverb_path
193
+
194
+ # Function to preprocess vocals only, i.e. apply just dereverb process.
195
+ def preprocess_vocals_only(song_input, mdx_model_params, song_id, is_webui, input_type, progress=None):
196
+ orig_song_path = song_input
197
+
198
+ song_output_dir = os.path.join(output_dir, song_id)
199
+ orig_song_path = convert_to_stereo(orig_song_path)
200
+
201
+ display_progress('[~] Applying DeReverb to Vocals...', 0.3, is_webui, progress)
202
+ _, 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)
203
+
204
+ return orig_song_path, vocals_path
205
+
206
+
207
+ 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):
208
+ rvc_model_path, rvc_index_path = get_rvc_model(voice_model, is_webui)
209
+ device = 'cpu'
210
+ config = Config(device, True)
211
+ hubert_model = load_hubert(device, config.is_half, os.path.join(rvc_models_dir, 'hubert_base.pt'))
212
+ cpt, version, net_g, tgt_sr, vc = get_vc(device, config.is_half, config, rvc_model_path)
213
+
214
+ # convert main vocals
215
+ rvc_infer(rvc_index_path, index_rate, vocals_path, output_path, pitch_change, f0_method, cpt, version, net_g, filter_radius, tgt_sr, rms_mix_rate, protect, crepe_hop_length, vc, hubert_model)
216
+ del hubert_model, cpt
217
+ gc.collect()
218
+
219
+
220
+ def add_audio_effects(audio_path, reverb_rm_size, reverb_wet, reverb_dry, reverb_damping):
221
+ # Added _covervocals at the end of filename
222
+ output_path = f'{os.path.splitext(audio_path)[0]}_mixed_covervocals.wav'
223
+
224
+ # Initialize audio effects plugins
225
+ board = Pedalboard(
226
+ [
227
+ HighpassFilter(),
228
+ Compressor(ratio=4, threshold_db=-15),
229
+ Reverb(room_size=reverb_rm_size, dry_level=reverb_dry, wet_level=reverb_wet, damping=reverb_damping)
230
+ ]
231
+ )
232
+
233
+ with AudioFile(audio_path) as f:
234
+ with AudioFile(output_path, 'w', f.samplerate, f.num_channels) as o:
235
+ # Read one second of audio at a time, until the file is empty:
236
+ while f.tell() < f.frames:
237
+ chunk = f.read(int(f.samplerate))
238
+ effected = board(chunk, f.samplerate, reset=False)
239
+ o.write(effected)
240
+
241
+ return output_path
242
+
243
+
244
+ def combine_audio(audio_paths, output_path, main_gain, backup_gain, inst_gain, output_format):
245
+ main_vocal_audio = AudioSegment.from_wav(audio_paths[0]) - 4 + main_gain
246
+ backup_vocal_audio = AudioSegment.from_wav(audio_paths[1]) - 6 + backup_gain
247
+ instrumental_audio = AudioSegment.from_wav(audio_paths[2]) - 7 + inst_gain
248
+ main_vocal_audio.overlay(backup_vocal_audio).overlay(instrumental_audio).export(output_path, format=output_format)
249
+
250
+ # Function defining the pipeline to create the AI cover of an audio containing only voice information.
251
+ # Returns the path of the cover vocal.
252
+ def vocal_only_pipeline(song_input, voice_model, pitch_change, is_webui=0,index_rate=0.5, filter_radius=3,
253
+ rms_mix_rate=0.25, f0_method='rmvpe', crepe_hop_length=128, protect=0.33, pitch_change_all=0,
254
+ reverb_rm_size=0.15, reverb_wet=0.2, reverb_dry=0.8, reverb_damping=0.7, output_format='mp3',progress=gr.Progress()):
255
+ try:
256
+ #Load mdx parameters
257
+ with open(os.path.join(mdxnet_models_dir, 'model_data.json')) as infile:
258
+ mdx_model_params = json.load(infile)
259
+
260
+ # Get path of loaded vocal file
261
+ input_type = 'local'
262
+ song_input = song_input.strip('\"')
263
+ if os.path.exists(song_input):
264
+ song_id = get_hash(song_input)
265
+ else:
266
+ error_msg = f'{song_input} does not exist.'
267
+ song_id = None
268
+ raise_exception(error_msg, is_webui)
269
+
270
+ song_dir = os.path.join(output_dir, song_id)
271
+ os.makedirs(song_dir)
272
+
273
+ if not os.path.exists(song_dir):
274
+ os.makedirs(song_dir)
275
+ orig_song_path, vocals_path = preprocess_vocals_only(song_input, mdx_model_params, song_id, is_webui, input_type, progress)
276
+ else:
277
+ vocals_path = None
278
+ for file in os.listdir(song_dir):
279
+ vocals_path = os.path.join(song_dir, file)
280
+ return vocals_path
281
+
282
+ # if any of the audio files aren't available or keep intermediate files, rerun preprocess
283
+ #if any(path is None for path in paths):
284
+ # orig_song_path, vocals_path= preprocess_vocals_only(song_input, mdx_model_params, song_id, is_webui, input_type, progress)
285
+ #else:
286
+ # orig_song_path = paths
287
+ #orig_song_path, vocals_path = preprocess_vocals_only(song_input, mdx_model_params, song_id, is_webui, input_type, progress)
288
+
289
+ pitch_change = pitch_change * 12 + pitch_change_all
290
+ 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')
291
+ display_progress(f'[~] Post reverb {ai_vocals_path}', 0.5, is_webui, progress)
292
+ time.sleep(10)
293
+
294
+ display_progress('[~] Converting voice using RVC...', 0.5, is_webui, progress)
295
+ 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)
296
+ display_progress('[~] Applying audio effects to Vocals...', 0.8, is_webui, progress)
297
+ ai_vocals_mixed_path = add_audio_effects(ai_vocals_path, reverb_rm_size, reverb_wet, reverb_dry, reverb_damping)
298
+
299
+ return ai_vocals_mixed_path
300
+ except Exception as e:
301
+ raise_exception(str(e), is_webui)
302
+
303
+
304
+
305
+ def song_cover_pipeline(song_input, voice_model, pitch_change, keep_files,
306
+ is_webui=0, main_gain=0, backup_gain=0, inst_gain=0, index_rate=0.5, filter_radius=3,
307
+ rms_mix_rate=0.25, f0_method='rmvpe', crepe_hop_length=128, protect=0.33, pitch_change_all=0,
308
+ reverb_rm_size=0.15, reverb_wet=0.2, reverb_dry=0.8, reverb_damping=0.7, output_format='mp3',
309
+ progress=gr.Progress()):
310
+
311
+ try:
312
+ if not song_input or not voice_model:
313
+ raise_exception('Ensure that the song input field and voice model field is filled.', is_webui)
314
+
315
+ display_progress('[~] Starting AI Cover Generation Pipeline...', 0, is_webui, progress)
316
+
317
+ with open(os.path.join(mdxnet_models_dir, 'model_data.json')) as infile:
318
+ mdx_model_params = json.load(infile)
319
+
320
+ # if youtube url
321
+ if urlparse(song_input).scheme == 'https':
322
+ input_type = 'yt'
323
+ song_id = get_youtube_video_id(song_input)
324
+ if song_id is None:
325
+ error_msg = 'Invalid YouTube url.'
326
+ raise_exception(error_msg, is_webui)
327
+
328
+ # local audio file
329
+ else:
330
+ input_type = 'local'
331
+ song_input = song_input.strip('\"')
332
+ if os.path.exists(song_input):
333
+ song_id = get_hash(song_input)
334
+ else:
335
+ error_msg = f'{song_input} does not exist.'
336
+ song_id = None
337
+ raise_exception(error_msg, is_webui)
338
+
339
+ song_dir = os.path.join(output_dir, song_id)
340
+
341
+ if not os.path.exists(song_dir):
342
+ os.makedirs(song_dir)
343
+ orig_song_path, vocals_path, instrumentals_path, main_vocals_path, backup_vocals_path, main_vocals_dereverb_path = preprocess_song(song_input, mdx_model_params, song_id, is_webui, input_type, progress)
344
+
345
+ else:
346
+ vocals_path, main_vocals_path = None, None
347
+ paths = get_audio_paths(song_dir)
348
+
349
+ # if any of the audio files aren't available or keep intermediate files, rerun preprocess
350
+ if any(path is None for path in paths) or keep_files:
351
+ orig_song_path, vocals_path, instrumentals_path, main_vocals_path, backup_vocals_path, main_vocals_dereverb_path = preprocess_song(song_input, mdx_model_params, song_id, is_webui, input_type, progress)
352
+ else:
353
+ orig_song_path, instrumentals_path, main_vocals_dereverb_path, backup_vocals_path = paths
354
+
355
+ pitch_change = pitch_change * 12 + pitch_change_all
356
+ 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')
357
+ # Added _cover at the end of filename
358
+ ai_cover_path = os.path.join(song_dir, f'{os.path.splitext(os.path.basename(orig_song_path))[0]} ({voice_model} Ver)_cover.{output_format}')
359
+
360
+ if not os.path.exists(ai_vocals_path):
361
+ display_progress('[~] Converting voice using RVC...', 0.5, is_webui, progress)
362
+ voice_change(voice_model, main_vocals_dereverb_path, ai_vocals_path, pitch_change, f0_method, index_rate, filter_radius, rms_mix_rate, protect, crepe_hop_length, is_webui)
363
+
364
+ display_progress('[~] Applying audio effects to Vocals...', 0.8, is_webui, progress)
365
+ ai_vocals_mixed_path = add_audio_effects(ai_vocals_path, reverb_rm_size, reverb_wet, reverb_dry, reverb_damping)
366
+
367
+ if pitch_change_all != 0:
368
+ display_progress('[~] Applying overall pitch change', 0.85, is_webui, progress)
369
+ instrumentals_path = pitch_shift(instrumentals_path, pitch_change_all)
370
+ backup_vocals_path = pitch_shift(backup_vocals_path, pitch_change_all)
371
+
372
+ display_progress('[~] Combining AI Vocals and Instrumentals...', 0.9, is_webui, progress)
373
+ combine_audio([ai_vocals_mixed_path, backup_vocals_path, instrumentals_path], ai_cover_path, main_gain, backup_gain, inst_gain, output_format)
374
+
375
+ if not keep_files:
376
+ display_progress('[~] Removing intermediate audio files...', 0.95, is_webui, progress)
377
+ intermediate_files = [vocals_path, main_vocals_path, ai_vocals_mixed_path]
378
+ if pitch_change_all != 0:
379
+ intermediate_files += [instrumentals_path, backup_vocals_path]
380
+ for file in intermediate_files:
381
+ if file and os.path.exists(file):
382
+ os.remove(file)
383
+
384
+
385
+ # Returning the stems: AI cover, original vocal, original instrumental, AI generated vocal
386
+
387
+ return ai_cover_path, vocals_path, instrumentals_path, ai_vocals_mixed_path
388
+
389
+ except Exception as e:
390
+ raise_exception(str(e), is_webui)
391
+
392
+
393
+ if __name__ == '__main__':
394
+ parser = argparse.ArgumentParser(description='Generate a AI cover song in the song_output/id directory.', add_help=True)
395
+ parser.add_argument('-i', '--song-input', type=str, required=True, help='Link to a YouTube video or the filepath to a local mp3/wav file to create an AI cover of')
396
+ parser.add_argument('-dir', '--rvc-dirname', type=str, required=True, help='Name of the folder in the rvc_models directory containing the RVC model file and optional index file to use')
397
+ parser.add_argument('-p', '--pitch-change', type=int, required=True, help='Change the pitch of AI Vocals only. Generally, use 1 for male to female and -1 for vice-versa. (Octaves)')
398
+ parser.add_argument('-k', '--keep-files', action=argparse.BooleanOptionalAction, help='Whether to keep all intermediate audio files generated in the song_output/id directory, e.g. Isolated Vocals/Instrumentals')
399
+ parser.add_argument('-ir', '--index-rate', type=float, default=0.5, help='A decimal number e.g. 0.5, used to reduce/resolve the timbre leakage problem. If set to 1, more biased towards the timbre quality of the training dataset')
400
+ parser.add_argument('-fr', '--filter-radius', type=int, default=3, help='A number between 0 and 7. If >=3: apply median filtering to the harvested pitch results. The value represents the filter radius and can reduce breathiness.')
401
+ parser.add_argument('-rms', '--rms-mix-rate', type=float, default=0.25, help="A decimal number e.g. 0.25. Control how much to use the original vocal's loudness (0) or a fixed loudness (1).")
402
+ parser.add_argument('-palgo', '--pitch-detection-algo', type=str, default='rmvpe', help='Best option is rmvpe (clarity in vocals), then mangio-crepe (smoother vocals).')
403
+ parser.add_argument('-hop', '--crepe-hop-length', type=int, default=128, help='If pitch detection algo is mangio-crepe, controls how often it checks for pitch changes in milliseconds. The higher the value, the faster the conversion and less risk of voice cracks, but there is less pitch accuracy. Recommended: 128.')
404
+ parser.add_argument('-pro', '--protect', type=float, default=0.33, help='A decimal number e.g. 0.33. Protect voiceless consonants and breath sounds to prevent artifacts such as tearing in electronic music. Set to 0.5 to disable. Decrease the value to increase protection, but it may reduce indexing accuracy.')
405
+ parser.add_argument('-mv', '--main-vol', type=int, default=0, help='Volume change for AI main vocals in decibels. Use -3 to decrease by 3 decibels and 3 to increase by 3 decibels')
406
+ parser.add_argument('-bv', '--backup-vol', type=int, default=0, help='Volume change for backup vocals in decibels')
407
+ parser.add_argument('-iv', '--inst-vol', type=int, default=0, help='Volume change for instrumentals in decibels')
408
+ parser.add_argument('-pall', '--pitch-change-all', type=int, default=0, help='Change the pitch/key of vocals and instrumentals. Changing this slightly reduces sound quality')
409
+ parser.add_argument('-rsize', '--reverb-size', type=float, default=0.15, help='Reverb room size between 0 and 1')
410
+ parser.add_argument('-rwet', '--reverb-wetness', type=float, default=0.2, help='Reverb wet level between 0 and 1')
411
+ parser.add_argument('-rdry', '--reverb-dryness', type=float, default=0.8, help='Reverb dry level between 0 and 1')
412
+ parser.add_argument('-rdamp', '--reverb-damping', type=float, default=0.7, help='Reverb damping between 0 and 1')
413
+ parser.add_argument('-oformat', '--output-format', type=str, default='mp3', help='Output format of audio file. mp3 for smaller file size, wav for best quality')
414
+ args = parser.parse_args()
415
+
416
+ rvc_dirname = args.rvc_dirname
417
+ if not os.path.exists(os.path.join(rvc_models_dir, rvc_dirname)):
418
+ raise Exception(f'The folder {os.path.join(rvc_models_dir, rvc_dirname)} does not exist.')
419
+
420
+ cover_path, original_vocals, original_instrumentals, ai_vocals= song_cover_pipeline(args.song_input, rvc_dirname, args.pitch_change, args.keep_files,
421
+ #cover_path, original_vocals = song_cover_pipeline(args.song_input, rvc_dirname, args.pitch_change, args.keep_files,
422
+ main_gain=args.main_vol, backup_gain=args.backup_vol, inst_gain=args.inst_vol,
423
+ index_rate=args.index_rate, filter_radius=args.filter_radius,
424
+ rms_mix_rate=args.rms_mix_rate, f0_method=args.pitch_detection_algo,
425
+ crepe_hop_length=args.crepe_hop_length, protect=args.protect,
426
+ pitch_change_all=args.pitch_change_all,
427
+ reverb_rm_size=args.reverb_size, reverb_wet=args.reverb_wetness,
428
+ reverb_dry=args.reverb_dryness, reverb_damping=args.reverb_damping,
429
+ output_format=args.output_format, vocal_only=False)
430
+
431
+
432
+
433
+ print(f'[+] Cover generated at {cover_path}')
434
+ print(f'[+] Original vocals at {original_vocals}')
435
+ print(f'[+] Original instrumentals at {original_instrumentals}')
436
+ print(f'[+] AI vocals at {ai_vocals}')
437
+
438
+