Update main/app/core/ui.py
Browse files- main/app/core/ui.py +361 -361
main/app/core/ui.py
CHANGED
|
@@ -1,362 +1,362 @@
|
|
| 1 |
-
import os
|
| 2 |
-
import re
|
| 3 |
-
import sys
|
| 4 |
-
import json
|
| 5 |
-
import torch
|
| 6 |
-
import shutil
|
| 7 |
-
|
| 8 |
-
import gradio as gr
|
| 9 |
-
import sounddevice as sd
|
| 10 |
-
|
| 11 |
-
sys.path.append(os.getcwd())
|
| 12 |
-
|
| 13 |
-
from main.library.backends import directml, opencl
|
| 14 |
-
from main.inference.realtime.audio import list_audio_device
|
| 15 |
-
from main.app.variables import config, configs, configs_json, logger, translations, edgetts, google_tts_voice, method_f0, method_f0_full, vr_models, mdx_models, demucs_models, embedders_model, spin_model, whisper_model
|
| 16 |
-
|
| 17 |
-
def gr_info(message):
|
| 18 |
-
gr.Info(message, duration=2)
|
| 19 |
-
logger.info(message)
|
| 20 |
-
|
| 21 |
-
def gr_warning(message):
|
| 22 |
-
gr.Warning(message, duration=2)
|
| 23 |
-
logger.warning(message)
|
| 24 |
-
|
| 25 |
-
def gr_error(message):
|
| 26 |
-
gr.Error(message=message, duration=6)
|
| 27 |
-
logger.error(message)
|
| 28 |
-
|
| 29 |
-
def get_gpu_info():
|
| 30 |
-
ngpu = torch.cuda.device_count()
|
| 31 |
-
gpu_infos = [
|
| 32 |
-
f"{i}: {torch.cuda.get_device_name(i)} ({int(torch.cuda.get_device_properties(i).total_memory / 1024 / 1024 / 1024 + 0.4)} GB)"
|
| 33 |
-
for i in range(ngpu)
|
| 34 |
-
if torch.cuda.is_available() or ngpu != 0
|
| 35 |
-
]
|
| 36 |
-
|
| 37 |
-
if len(gpu_infos) == 0:
|
| 38 |
-
if directml.torch_available:
|
| 39 |
-
ngpu = directml.device_count()
|
| 40 |
-
gpu_infos = [f"{i}: {directml.device_name(i)}" for i in range(ngpu) if directml.is_available() or ngpu != 0]
|
| 41 |
-
elif opencl.torch_available:
|
| 42 |
-
ngpu = opencl.device_count()
|
| 43 |
-
gpu_infos = [f"{i}: {opencl.device_name(i)}" for i in range(ngpu) if opencl.is_available() or ngpu != 0]
|
| 44 |
-
else:
|
| 45 |
-
ngpu = 0
|
| 46 |
-
gpu_infos = []
|
| 47 |
-
|
| 48 |
-
return "\n".join(gpu_infos) if len(gpu_infos) > 0 and not config.cpu_mode else translations["no_support_gpu"]
|
| 49 |
-
|
| 50 |
-
def gpu_number_str():
|
| 51 |
-
if config.cpu_mode: return "-"
|
| 52 |
-
|
| 53 |
-
ngpu = torch.cuda.device_count()
|
| 54 |
-
if ngpu == 0: ngpu = directml.device_count() if directml.torch_available else opencl.device_count()
|
| 55 |
-
|
| 56 |
-
return str("-".join(map(str, range(ngpu))) if torch.cuda.is_available() or directml.is_available() or opencl.is_available() else "-")
|
| 57 |
-
|
| 58 |
-
def change_f0_choices():
|
| 59 |
-
f0_file = sorted([os.path.abspath(os.path.join(root, f)) for root, _, files in os.walk(configs["f0_path"]) for f in files if f.endswith(".txt")])
|
| 60 |
-
return {"value": f0_file[0] if len(f0_file) >= 1 else "", "choices": f0_file, "__type__": "update"}
|
| 61 |
-
|
| 62 |
-
def change_audios_choices(input_audio):
|
| 63 |
-
audios = sorted([os.path.abspath(os.path.join(root, f)) for root, _, files in os.walk(configs["audios_path"]) for f in files if os.path.splitext(f)[1].lower() in (".wav", ".mp3", ".flac", ".ogg", ".opus", ".m4a", ".mp4", ".aac", ".alac", ".wma", ".aiff", ".webm", ".ac3")])
|
| 64 |
-
return {"value": input_audio if input_audio != "" else (audios[0] if len(audios) >= 1 else ""), "choices": audios, "__type__": "update"}
|
| 65 |
-
|
| 66 |
-
def change_reference_choices():
|
| 67 |
-
reference = sorted([re.sub(r'_v\d+_(?:[A-Za-z0-9_]+?)_(True|False)_(True|False)$', '', name) for name in os.listdir(configs["reference_path"]) if os.path.exists(os.path.join(configs["reference_path"], name)) and os.path.isdir(os.path.join(configs["reference_path"], name))])
|
| 68 |
-
return {"value": reference[0] if len(reference) >= 1 else "", "choices": reference, "__type__": "update"}
|
| 69 |
-
|
| 70 |
-
def change_models_choices():
|
| 71 |
-
model, index = sorted(list(model for model in os.listdir(configs["weights_path"]) if model.endswith((".pth", ".onnx")) and not model.startswith("G_") and not model.startswith("D_"))), sorted([os.path.join(root, name) for root, _, files in os.walk(configs["logs_path"], topdown=False) for name in files if name.endswith(".index") and "trained" not in name])
|
| 72 |
-
return [{"value": model[0] if len(model) >= 1 else "", "choices": model, "__type__": "update"}, {"value": index[0] if len(index) >= 1 else "", "choices": index, "__type__": "update"}]
|
| 73 |
-
|
| 74 |
-
def change_pretrained_choices():
|
| 75 |
-
pretrainD = sorted([model for model in os.listdir(configs["pretrained_custom_path"]) if model.endswith(".pth") and "D" in model])
|
| 76 |
-
pretrainG = sorted([model for model in os.listdir(configs["pretrained_custom_path"]) if model.endswith(".pth") and "G" in model])
|
| 77 |
-
|
| 78 |
-
return [{"choices": pretrainD, "value": pretrainD[0] if len(pretrainD) >= 1 else "", "__type__": "update"}, {"choices": pretrainG, "value": pretrainG[0] if len(pretrainG) >= 1 else "", "__type__": "update"}]
|
| 79 |
-
|
| 80 |
-
def change_choices_del():
|
| 81 |
-
return [{"choices": sorted(list(model for model in os.listdir(configs["weights_path"]) if model.endswith(".pth") and not model.startswith("G_") and not model.startswith("D_"))), "__type__": "update"}, {"choices": sorted([os.path.join(configs["logs_path"], f) for f in os.listdir(configs["logs_path"]) if f not in ["mute", "reference"] and os.path.isdir(os.path.join(configs["logs_path"], f))]), "__type__": "update"}]
|
| 82 |
-
|
| 83 |
-
def change_preset_choices():
|
| 84 |
-
return {"value": "", "choices": sorted(list(f for f in os.listdir(configs["presets_path"]) if f.endswith(".conversion.json"))), "__type__": "update"}
|
| 85 |
-
|
| 86 |
-
def change_effect_preset_choices():
|
| 87 |
-
return {"value": "", "choices": sorted(list(f for f in os.listdir(configs["presets_path"]) if f.endswith(".effect.json"))), "__type__": "update"}
|
| 88 |
-
|
| 89 |
-
def change_tts_voice_choices(google):
|
| 90 |
-
return {"choices": google_tts_voice if google else edgetts, "value": google_tts_voice[0] if google else edgetts[0], "__type__": "update"}
|
| 91 |
-
|
| 92 |
-
def change_backing_choices(backing, merge):
|
| 93 |
-
if backing or merge: return {"value": False, "interactive": False, "__type__": "update"}
|
| 94 |
-
elif not backing or not merge: return {"interactive": True, "__type__": "update"}
|
| 95 |
-
else: gr_warning(translations["option_not_valid"])
|
| 96 |
-
|
| 97 |
-
def change_download_choices(select):
|
| 98 |
-
selects = [False]*10
|
| 99 |
-
|
| 100 |
-
if select == translations["download_url"]: selects[0] = selects[1] = selects[2] = True
|
| 101 |
-
elif select == translations["download_from_csv"]: selects[3] = selects[4] = True
|
| 102 |
-
elif select == translations["search_models"]: selects[5] = selects[6] = True
|
| 103 |
-
elif select == translations["upload"]: selects[9] = True
|
| 104 |
-
else: gr_warning(translations["option_not_valid"])
|
| 105 |
-
|
| 106 |
-
return [{"visible": selects[i], "__type__": "update"} for i in range(len(selects))]
|
| 107 |
-
|
| 108 |
-
def change_download_pretrained_choices(select):
|
| 109 |
-
selects = [False]*7
|
| 110 |
-
|
| 111 |
-
if select == translations["download_url"]: selects[0] = selects[1] = selects[2] = True
|
| 112 |
-
elif select == translations["list_model"]: selects[3] = selects[4] = selects[5] = True
|
| 113 |
-
elif select == translations["upload"]: selects[6] = True
|
| 114 |
-
else: gr_warning(translations["option_not_valid"])
|
| 115 |
-
|
| 116 |
-
return [{"visible": selects[i], "__type__": "update"} for i in range(len(selects))]
|
| 117 |
-
|
| 118 |
-
def get_index(model):
|
| 119 |
-
model = os.path.basename(model).split("_")[0]
|
| 120 |
-
return {"value": next((f for f in [os.path.join(root, name) for root, _, files in os.walk(configs["logs_path"], topdown=False) for name in files if name.endswith(".index") and "trained" not in name] if model.split(".")[0] in f), ""), "__type__": "update"} if model else None
|
| 121 |
-
|
| 122 |
-
def index_strength_show(index):
|
| 123 |
-
return {"visible": index != "" and index != None and os.path.exists(index) and os.path.isfile(index), "value": 0.5, "__type__": "update"}
|
| 124 |
-
|
| 125 |
-
def hoplength_show(method, hybrid_method=None):
|
| 126 |
-
visible = False
|
| 127 |
-
|
| 128 |
-
for m in ["mangio-crepe", "fcpe", "yin", "piptrack", "mangio-penn"]:
|
| 129 |
-
if m in method: visible = True
|
| 130 |
-
if hybrid_method is not None and m in hybrid_method: visible = True
|
| 131 |
-
|
| 132 |
-
if visible: break
|
| 133 |
-
else: visible = False
|
| 134 |
-
|
| 135 |
-
return {"visible": visible, "__type__": "update"}
|
| 136 |
-
|
| 137 |
-
def visible(value):
|
| 138 |
-
return {"visible": value, "__type__": "update"}
|
| 139 |
-
|
| 140 |
-
def valueFalse_interactive(value):
|
| 141 |
-
return {"value": False, "interactive": value, "__type__": "update"}
|
| 142 |
-
|
| 143 |
-
def valueEmpty_visible1(value):
|
| 144 |
-
return {"value": "", "visible": value, "__type__": "update"}
|
| 145 |
-
|
| 146 |
-
def pitch_guidance_lock(vocoders):
|
| 147 |
-
return {"value": True, "interactive": vocoders == "Default", "__type__": "update"}
|
| 148 |
-
|
| 149 |
-
def vocoders_lock(pitch, vocoders):
|
| 150 |
-
return {"value": vocoders if pitch else "Default", "interactive": pitch, "__type__": "update"}
|
| 151 |
-
|
| 152 |
-
def unlock_f0(value):
|
| 153 |
-
return {"choices": method_f0_full if value else method_f0, "value": "rmvpe", "__type__": "update"}
|
| 154 |
-
|
| 155 |
-
def unlock_vocoder(value, vocoder):
|
| 156 |
-
return {"value": vocoder if value == "v2" else "Default", "interactive": value == "v2", "__type__": "update"}
|
| 157 |
-
|
| 158 |
-
def unlock_ver(value, vocoder):
|
| 159 |
-
return {"value": "v2" if vocoder == "Default" else value, "interactive": vocoder == "Default", "__type__": "update"}
|
| 160 |
-
|
| 161 |
-
def change_embedders_mode(value):
|
| 162 |
-
if value == "spin":
|
| 163 |
-
return {"value": spin_model[0], "choices": spin_model, "__type__": "update"}
|
| 164 |
-
elif value == "whisper":
|
| 165 |
-
return {"value": whisper_model[0], "choices": whisper_model, "__type__": "update"}
|
| 166 |
-
else:
|
| 167 |
-
return {"value": embedders_model[0], "choices": embedders_model, "__type__": "update"}
|
| 168 |
-
|
| 169 |
-
def change_fp(fp):
|
| 170 |
-
fp16 = fp == "fp16"
|
| 171 |
-
|
| 172 |
-
if fp16 and config.device in ["cpu", "mps", "ocl:0"]:
|
| 173 |
-
gr_warning(translations["fp16_not_support"])
|
| 174 |
-
return "fp32"
|
| 175 |
-
else:
|
| 176 |
-
gr_info(translations["start_update_precision"])
|
| 177 |
-
|
| 178 |
-
configs = json.load(open(configs_json, "r"))
|
| 179 |
-
configs["fp16"] = config.is_half = fp16
|
| 180 |
-
|
| 181 |
-
with open(configs_json, "w") as f:
|
| 182 |
-
json.dump(configs, f, indent=4)
|
| 183 |
-
|
| 184 |
-
gr_info(translations["success"])
|
| 185 |
-
return "fp16" if fp16 else "fp32"
|
| 186 |
-
|
| 187 |
-
def process_output(file_path):
|
| 188 |
-
if config.configs.get("delete_exists_file", True):
|
| 189 |
-
if os.path.exists(file_path) and os.path.isfile(file_path): os.remove(file_path)
|
| 190 |
-
return file_path
|
| 191 |
-
else:
|
| 192 |
-
if not os.path.exists(file_path): return file_path
|
| 193 |
-
file = os.path.splitext(os.path.basename(file_path))
|
| 194 |
-
|
| 195 |
-
index = 1
|
| 196 |
-
while 1:
|
| 197 |
-
file_path = os.path.join(os.path.dirname(file_path), f"{file[0]}_{index}{file[1]}")
|
| 198 |
-
if not os.path.exists(file_path): return file_path
|
| 199 |
-
index += 1
|
| 200 |
-
|
| 201 |
-
def shutil_move(input_path, output_path):
|
| 202 |
-
output_path = os.path.join(output_path, os.path.basename(input_path)) if os.path.isdir(output_path) else output_path
|
| 203 |
-
|
| 204 |
-
return shutil.move(input_path, process_output(output_path)) if os.path.exists(output_path) else shutil.move(input_path, output_path)
|
| 205 |
-
|
| 206 |
-
def separate_change(model_name, karaoke_model, reverb_model, enable_post_process, separate_backing, separate_reverb, enable_denoise):
|
| 207 |
-
model_type = "vr" if model_name in list(vr_models.keys()) else "mdx" if model_name in list(mdx_models.keys()) else "demucs" if model_name in list(demucs_models.keys()) else ""
|
| 208 |
-
karaoke_type = ("vr" if karaoke_model.startswith("VR") else "mdx") if separate_backing else None
|
| 209 |
-
reverb_type = ("vr" if not reverb_model.startswith("MDX") else "mdx") if separate_reverb else None
|
| 210 |
-
|
| 211 |
-
all_types = {model_type, karaoke_type, reverb_type}
|
| 212 |
-
|
| 213 |
-
is_vr = "vr" in all_types
|
| 214 |
-
is_mdx = "mdx" in all_types
|
| 215 |
-
is_demucs = "demucs" in all_types
|
| 216 |
-
|
| 217 |
-
return [
|
| 218 |
-
visible(separate_backing),
|
| 219 |
-
visible(separate_reverb),
|
| 220 |
-
visible(is_mdx or is_demucs),
|
| 221 |
-
visible(is_mdx or is_demucs),
|
| 222 |
-
visible(is_mdx),
|
| 223 |
-
visible(is_mdx or is_vr),
|
| 224 |
-
visible(is_demucs),
|
| 225 |
-
visible(is_vr),
|
| 226 |
-
visible(is_vr),
|
| 227 |
-
visible(is_vr and enable_post_process),
|
| 228 |
-
visible(is_vr and enable_denoise),
|
| 229 |
-
valueFalse_interactive(is_vr),
|
| 230 |
-
valueFalse_interactive(is_vr),
|
| 231 |
-
valueFalse_interactive(is_vr)
|
| 232 |
-
]
|
| 233 |
-
|
| 234 |
-
def create_dataset_change(model_name, reverb_model, enable_post_process, separate_reverb, enable_denoise):
|
| 235 |
-
model_type = "vr" if model_name in list(vr_models.keys()) else "mdx" if model_name in list(mdx_models.keys()) else "demucs" if model_name in list(demucs_models.keys()) else ""
|
| 236 |
-
reverb_type = ("vr" if not reverb_model.startswith("MDX") else "mdx") if separate_reverb else None
|
| 237 |
-
all_types = {model_type, reverb_type}
|
| 238 |
-
|
| 239 |
-
is_vr = "vr" in all_types
|
| 240 |
-
is_mdx = "mdx" in all_types
|
| 241 |
-
is_demucs = "demucs" in all_types
|
| 242 |
-
|
| 243 |
-
return [
|
| 244 |
-
visible(separate_reverb),
|
| 245 |
-
visible(is_mdx or is_demucs),
|
| 246 |
-
visible(is_mdx or is_demucs),
|
| 247 |
-
visible(is_mdx),
|
| 248 |
-
visible(is_mdx or is_vr),
|
| 249 |
-
visible(is_demucs),
|
| 250 |
-
visible(is_vr),
|
| 251 |
-
visible(is_vr),
|
| 252 |
-
visible(is_vr and enable_post_process),
|
| 253 |
-
visible(is_vr and enable_denoise),
|
| 254 |
-
valueFalse_interactive(is_vr),
|
| 255 |
-
valueFalse_interactive(is_vr),
|
| 256 |
-
valueFalse_interactive(is_vr)
|
| 257 |
-
]
|
| 258 |
-
|
| 259 |
-
def audio_device():
|
| 260 |
-
try:
|
| 261 |
-
input_devices, output_devices = list_audio_device()
|
| 262 |
-
|
| 263 |
-
def priority(name):
|
| 264 |
-
n = name.lower()
|
| 265 |
-
if "virtual" in n:
|
| 266 |
-
return 0
|
| 267 |
-
if "vb" in n:
|
| 268 |
-
return 1
|
| 269 |
-
return 2
|
| 270 |
-
|
| 271 |
-
output_sorted = sorted(output_devices, key=lambda d: priority(d.name))
|
| 272 |
-
input_sorted = sorted(
|
| 273 |
-
input_devices, key=lambda d: priority(d.name), reverse=True
|
| 274 |
-
)
|
| 275 |
-
|
| 276 |
-
input_device_list = {
|
| 277 |
-
f"{input_sorted.index(d)+1}: {d.name} ({d.host_api})": [d.index, d.max_input_channels] for d in input_sorted
|
| 278 |
-
}
|
| 279 |
-
output_device_list = {
|
| 280 |
-
f"{output_sorted.index(d)+1}: {d.name} ({d.host_api})": [d.index, d.max_output_channels] for d in output_sorted
|
| 281 |
-
}
|
| 282 |
-
|
| 283 |
-
return input_device_list, output_device_list
|
| 284 |
-
except Exception:
|
| 285 |
-
return [], []
|
| 286 |
-
|
| 287 |
-
def update_audio_device(input_device, output_device, monitor_device, monitor):
|
| 288 |
-
input_channels_map, output_channels_map = audio_device()
|
| 289 |
-
|
| 290 |
-
input_is_asio = "ASIO" in input_device if input_device else False
|
| 291 |
-
output_is_asio = "ASIO" in output_device if output_device else False
|
| 292 |
-
monitor_is_asio = "ASIO" in monitor_device if monitor_device else False
|
| 293 |
-
|
| 294 |
-
try:
|
| 295 |
-
input_max_ch = input_channels_map.get(input_device, [])[1]
|
| 296 |
-
output_max_ch = output_channels_map.get(output_device, [])[1]
|
| 297 |
-
monitor_max_ch = output_channels_map.get(monitor_device, [])[1] if monitor else 128
|
| 298 |
-
except:
|
| 299 |
-
input_max_ch = output_max_ch = monitor_max_ch = -1
|
| 300 |
-
|
| 301 |
-
return [
|
| 302 |
-
visible(monitor),
|
| 303 |
-
visible(monitor),
|
| 304 |
-
visible(monitor_is_asio),
|
| 305 |
-
visible(input_is_asio or output_is_asio or monitor_is_asio),
|
| 306 |
-
gr.update(visible=input_is_asio, maximum=input_max_ch),
|
| 307 |
-
gr.update(visible=output_is_asio, maximum=output_max_ch),
|
| 308 |
-
gr.update(visible=monitor_is_asio, maximum=monitor_max_ch)
|
| 309 |
-
]
|
| 310 |
-
|
| 311 |
-
def change_audio_device_choices():
|
| 312 |
-
sd._terminate()
|
| 313 |
-
sd._initialize()
|
| 314 |
-
|
| 315 |
-
input_channels_map, output_channels_map = audio_device()
|
| 316 |
-
input_channels_map, output_channels_map = list(input_channels_map.keys()), list(output_channels_map.keys())
|
| 317 |
-
|
| 318 |
-
return [
|
| 319 |
-
{"value": input_channels_map[0] if len(input_channels_map) >= 1 else "", "choices": input_channels_map, "__type__": "update"},
|
| 320 |
-
{"value": output_channels_map[0] if len(output_channels_map) >= 1 else "", "choices": output_channels_map, "__type__": "update"},
|
| 321 |
-
{"value": output_channels_map[0] if len(output_channels_map) >= 1 else "", "choices": output_channels_map, "__type__": "update"}
|
| 322 |
-
]
|
| 323 |
-
|
| 324 |
-
def replace_punctuation(filename):
|
| 325 |
-
return filename.replace(" ", "_").replace("-", "").replace("(", "").replace(")", "").replace("[", "").replace("]", "").replace(",", "").replace('"', "").replace("'", "").replace("|", "_").replace("{", "").replace("}", "").replace("-_-", "_").replace("_-_", "_").replace("-", "_").replace("---", "_").replace("___", "_").strip()
|
| 326 |
-
|
| 327 |
-
def replace_url(url):
|
| 328 |
-
return url.replace("/blob/", "/resolve/").replace("?download=true", "").strip()
|
| 329 |
-
|
| 330 |
-
def replace_modelname(modelname):
|
| 331 |
-
return replace_punctuation(modelname.replace(".onnx", "").replace(".pth", "").replace(".index", "").replace(".zip", ""))
|
| 332 |
-
|
| 333 |
-
def replace_export_format(audio_path, export_format = "wav"):
|
| 334 |
-
export_format = f".{export_format}"
|
| 335 |
-
|
| 336 |
-
return audio_path if audio_path.endswith(export_format) else audio_path.replace(f".{os.path.basename(audio_path).split('.')[-1]}", export_format)
|
| 337 |
-
|
| 338 |
-
def update_dropdowns_from_json(data):
|
| 339 |
-
if not data:
|
| 340 |
-
return [
|
| 341 |
-
gr.update(choices=[], value=None),
|
| 342 |
-
gr.update(choices=[], value=None),
|
| 343 |
-
gr.update(choices=[], value=None)
|
| 344 |
-
]
|
| 345 |
-
|
| 346 |
-
inputs =
|
| 347 |
-
outputs =
|
| 348 |
-
|
| 349 |
-
return [
|
| 350 |
-
gr.update(choices=inputs, value=None),
|
| 351 |
-
gr.update(choices=outputs, value=None),
|
| 352 |
-
gr.update(choices=outputs, value=None),
|
| 353 |
-
]
|
| 354 |
-
|
| 355 |
-
def update_button_from_json(data):
|
| 356 |
-
if not data:
|
| 357 |
-
return [gr.update(interactive=True), gr.update(interactive=False)]
|
| 358 |
-
|
| 359 |
-
return [
|
| 360 |
-
gr.update(interactive=data.get("start_button", True)),
|
| 361 |
-
gr.update(interactive=data.get("stop_button", False))
|
| 362 |
]
|
|
|
|
| 1 |
+
import os
|
| 2 |
+
import re
|
| 3 |
+
import sys
|
| 4 |
+
import json
|
| 5 |
+
import torch
|
| 6 |
+
import shutil
|
| 7 |
+
|
| 8 |
+
import gradio as gr
|
| 9 |
+
import sounddevice as sd
|
| 10 |
+
|
| 11 |
+
sys.path.append(os.getcwd())
|
| 12 |
+
|
| 13 |
+
from main.library.backends import directml, opencl
|
| 14 |
+
from main.inference.realtime.audio import list_audio_device
|
| 15 |
+
from main.app.variables import config, configs, configs_json, logger, translations, edgetts, google_tts_voice, method_f0, method_f0_full, vr_models, mdx_models, demucs_models, embedders_model, spin_model, whisper_model
|
| 16 |
+
|
| 17 |
+
def gr_info(message):
|
| 18 |
+
gr.Info(message, duration=2)
|
| 19 |
+
logger.info(message)
|
| 20 |
+
|
| 21 |
+
def gr_warning(message):
|
| 22 |
+
gr.Warning(message, duration=2)
|
| 23 |
+
logger.warning(message)
|
| 24 |
+
|
| 25 |
+
def gr_error(message):
|
| 26 |
+
gr.Error(message=message, duration=6)
|
| 27 |
+
logger.error(message)
|
| 28 |
+
|
| 29 |
+
def get_gpu_info():
|
| 30 |
+
ngpu = torch.cuda.device_count()
|
| 31 |
+
gpu_infos = [
|
| 32 |
+
f"{i}: {torch.cuda.get_device_name(i)} ({int(torch.cuda.get_device_properties(i).total_memory / 1024 / 1024 / 1024 + 0.4)} GB)"
|
| 33 |
+
for i in range(ngpu)
|
| 34 |
+
if torch.cuda.is_available() or ngpu != 0
|
| 35 |
+
]
|
| 36 |
+
|
| 37 |
+
if len(gpu_infos) == 0:
|
| 38 |
+
if directml.torch_available:
|
| 39 |
+
ngpu = directml.device_count()
|
| 40 |
+
gpu_infos = [f"{i}: {directml.device_name(i)}" for i in range(ngpu) if directml.is_available() or ngpu != 0]
|
| 41 |
+
elif opencl.torch_available:
|
| 42 |
+
ngpu = opencl.device_count()
|
| 43 |
+
gpu_infos = [f"{i}: {opencl.device_name(i)}" for i in range(ngpu) if opencl.is_available() or ngpu != 0]
|
| 44 |
+
else:
|
| 45 |
+
ngpu = 0
|
| 46 |
+
gpu_infos = []
|
| 47 |
+
|
| 48 |
+
return "\n".join(gpu_infos) if len(gpu_infos) > 0 and not config.cpu_mode else translations["no_support_gpu"]
|
| 49 |
+
|
| 50 |
+
def gpu_number_str():
|
| 51 |
+
if config.cpu_mode: return "-"
|
| 52 |
+
|
| 53 |
+
ngpu = torch.cuda.device_count()
|
| 54 |
+
if ngpu == 0: ngpu = directml.device_count() if directml.torch_available else opencl.device_count()
|
| 55 |
+
|
| 56 |
+
return str("-".join(map(str, range(ngpu))) if torch.cuda.is_available() or directml.is_available() or opencl.is_available() else "-")
|
| 57 |
+
|
| 58 |
+
def change_f0_choices():
|
| 59 |
+
f0_file = sorted([os.path.abspath(os.path.join(root, f)) for root, _, files in os.walk(configs["f0_path"]) for f in files if f.endswith(".txt")])
|
| 60 |
+
return {"value": f0_file[0] if len(f0_file) >= 1 else "", "choices": f0_file, "__type__": "update"}
|
| 61 |
+
|
| 62 |
+
def change_audios_choices(input_audio):
|
| 63 |
+
audios = sorted([os.path.abspath(os.path.join(root, f)) for root, _, files in os.walk(configs["audios_path"]) for f in files if os.path.splitext(f)[1].lower() in (".wav", ".mp3", ".flac", ".ogg", ".opus", ".m4a", ".mp4", ".aac", ".alac", ".wma", ".aiff", ".webm", ".ac3")])
|
| 64 |
+
return {"value": input_audio if input_audio != "" else (audios[0] if len(audios) >= 1 else ""), "choices": audios, "__type__": "update"}
|
| 65 |
+
|
| 66 |
+
def change_reference_choices():
|
| 67 |
+
reference = sorted([re.sub(r'_v\d+_(?:[A-Za-z0-9_]+?)_(True|False)_(True|False)$', '', name) for name in os.listdir(configs["reference_path"]) if os.path.exists(os.path.join(configs["reference_path"], name)) and os.path.isdir(os.path.join(configs["reference_path"], name))])
|
| 68 |
+
return {"value": reference[0] if len(reference) >= 1 else "", "choices": reference, "__type__": "update"}
|
| 69 |
+
|
| 70 |
+
def change_models_choices():
|
| 71 |
+
model, index = sorted(list(model for model in os.listdir(configs["weights_path"]) if model.endswith((".pth", ".onnx")) and not model.startswith("G_") and not model.startswith("D_"))), sorted([os.path.join(root, name) for root, _, files in os.walk(configs["logs_path"], topdown=False) for name in files if name.endswith(".index") and "trained" not in name])
|
| 72 |
+
return [{"value": model[0] if len(model) >= 1 else "", "choices": model, "__type__": "update"}, {"value": index[0] if len(index) >= 1 else "", "choices": index, "__type__": "update"}]
|
| 73 |
+
|
| 74 |
+
def change_pretrained_choices():
|
| 75 |
+
pretrainD = sorted([model for model in os.listdir(configs["pretrained_custom_path"]) if model.endswith(".pth") and "D" in model])
|
| 76 |
+
pretrainG = sorted([model for model in os.listdir(configs["pretrained_custom_path"]) if model.endswith(".pth") and "G" in model])
|
| 77 |
+
|
| 78 |
+
return [{"choices": pretrainD, "value": pretrainD[0] if len(pretrainD) >= 1 else "", "__type__": "update"}, {"choices": pretrainG, "value": pretrainG[0] if len(pretrainG) >= 1 else "", "__type__": "update"}]
|
| 79 |
+
|
| 80 |
+
def change_choices_del():
|
| 81 |
+
return [{"choices": sorted(list(model for model in os.listdir(configs["weights_path"]) if model.endswith(".pth") and not model.startswith("G_") and not model.startswith("D_"))), "__type__": "update"}, {"choices": sorted([os.path.join(configs["logs_path"], f) for f in os.listdir(configs["logs_path"]) if f not in ["mute", "reference"] and os.path.isdir(os.path.join(configs["logs_path"], f))]), "__type__": "update"}]
|
| 82 |
+
|
| 83 |
+
def change_preset_choices():
|
| 84 |
+
return {"value": "", "choices": sorted(list(f for f in os.listdir(configs["presets_path"]) if f.endswith(".conversion.json"))), "__type__": "update"}
|
| 85 |
+
|
| 86 |
+
def change_effect_preset_choices():
|
| 87 |
+
return {"value": "", "choices": sorted(list(f for f in os.listdir(configs["presets_path"]) if f.endswith(".effect.json"))), "__type__": "update"}
|
| 88 |
+
|
| 89 |
+
def change_tts_voice_choices(google):
|
| 90 |
+
return {"choices": google_tts_voice if google else edgetts, "value": google_tts_voice[0] if google else edgetts[0], "__type__": "update"}
|
| 91 |
+
|
| 92 |
+
def change_backing_choices(backing, merge):
|
| 93 |
+
if backing or merge: return {"value": False, "interactive": False, "__type__": "update"}
|
| 94 |
+
elif not backing or not merge: return {"interactive": True, "__type__": "update"}
|
| 95 |
+
else: gr_warning(translations["option_not_valid"])
|
| 96 |
+
|
| 97 |
+
def change_download_choices(select):
|
| 98 |
+
selects = [False]*10
|
| 99 |
+
|
| 100 |
+
if select == translations["download_url"]: selects[0] = selects[1] = selects[2] = True
|
| 101 |
+
elif select == translations["download_from_csv"]: selects[3] = selects[4] = True
|
| 102 |
+
elif select == translations["search_models"]: selects[5] = selects[6] = True
|
| 103 |
+
elif select == translations["upload"]: selects[9] = True
|
| 104 |
+
else: gr_warning(translations["option_not_valid"])
|
| 105 |
+
|
| 106 |
+
return [{"visible": selects[i], "__type__": "update"} for i in range(len(selects))]
|
| 107 |
+
|
| 108 |
+
def change_download_pretrained_choices(select):
|
| 109 |
+
selects = [False]*7
|
| 110 |
+
|
| 111 |
+
if select == translations["download_url"]: selects[0] = selects[1] = selects[2] = True
|
| 112 |
+
elif select == translations["list_model"]: selects[3] = selects[4] = selects[5] = True
|
| 113 |
+
elif select == translations["upload"]: selects[6] = True
|
| 114 |
+
else: gr_warning(translations["option_not_valid"])
|
| 115 |
+
|
| 116 |
+
return [{"visible": selects[i], "__type__": "update"} for i in range(len(selects))]
|
| 117 |
+
|
| 118 |
+
def get_index(model):
|
| 119 |
+
model = os.path.basename(model).split("_")[0]
|
| 120 |
+
return {"value": next((f for f in [os.path.join(root, name) for root, _, files in os.walk(configs["logs_path"], topdown=False) for name in files if name.endswith(".index") and "trained" not in name] if model.split(".")[0] in f), ""), "__type__": "update"} if model else None
|
| 121 |
+
|
| 122 |
+
def index_strength_show(index):
|
| 123 |
+
return {"visible": index != "" and index != None and os.path.exists(index) and os.path.isfile(index), "value": 0.5, "__type__": "update"}
|
| 124 |
+
|
| 125 |
+
def hoplength_show(method, hybrid_method=None):
|
| 126 |
+
visible = False
|
| 127 |
+
|
| 128 |
+
for m in ["mangio-crepe", "fcpe", "yin", "piptrack", "mangio-penn"]:
|
| 129 |
+
if m in method: visible = True
|
| 130 |
+
if hybrid_method is not None and m in hybrid_method: visible = True
|
| 131 |
+
|
| 132 |
+
if visible: break
|
| 133 |
+
else: visible = False
|
| 134 |
+
|
| 135 |
+
return {"visible": visible, "__type__": "update"}
|
| 136 |
+
|
| 137 |
+
def visible(value):
|
| 138 |
+
return {"visible": value, "__type__": "update"}
|
| 139 |
+
|
| 140 |
+
def valueFalse_interactive(value):
|
| 141 |
+
return {"value": False, "interactive": value, "__type__": "update"}
|
| 142 |
+
|
| 143 |
+
def valueEmpty_visible1(value):
|
| 144 |
+
return {"value": "", "visible": value, "__type__": "update"}
|
| 145 |
+
|
| 146 |
+
def pitch_guidance_lock(vocoders):
|
| 147 |
+
return {"value": True, "interactive": vocoders == "Default", "__type__": "update"}
|
| 148 |
+
|
| 149 |
+
def vocoders_lock(pitch, vocoders):
|
| 150 |
+
return {"value": vocoders if pitch else "Default", "interactive": pitch, "__type__": "update"}
|
| 151 |
+
|
| 152 |
+
def unlock_f0(value):
|
| 153 |
+
return {"choices": method_f0_full if value else method_f0, "value": "rmvpe", "__type__": "update"}
|
| 154 |
+
|
| 155 |
+
def unlock_vocoder(value, vocoder):
|
| 156 |
+
return {"value": vocoder if value == "v2" else "Default", "interactive": value == "v2", "__type__": "update"}
|
| 157 |
+
|
| 158 |
+
def unlock_ver(value, vocoder):
|
| 159 |
+
return {"value": "v2" if vocoder == "Default" else value, "interactive": vocoder == "Default", "__type__": "update"}
|
| 160 |
+
|
| 161 |
+
def change_embedders_mode(value):
|
| 162 |
+
if value == "spin":
|
| 163 |
+
return {"value": spin_model[0], "choices": spin_model, "__type__": "update"}
|
| 164 |
+
elif value == "whisper":
|
| 165 |
+
return {"value": whisper_model[0], "choices": whisper_model, "__type__": "update"}
|
| 166 |
+
else:
|
| 167 |
+
return {"value": embedders_model[0], "choices": embedders_model, "__type__": "update"}
|
| 168 |
+
|
| 169 |
+
def change_fp(fp):
|
| 170 |
+
fp16 = fp == "fp16"
|
| 171 |
+
|
| 172 |
+
if fp16 and config.device in ["cpu", "mps", "ocl:0"]:
|
| 173 |
+
gr_warning(translations["fp16_not_support"])
|
| 174 |
+
return "fp32"
|
| 175 |
+
else:
|
| 176 |
+
gr_info(translations["start_update_precision"])
|
| 177 |
+
|
| 178 |
+
configs = json.load(open(configs_json, "r"))
|
| 179 |
+
configs["fp16"] = config.is_half = fp16
|
| 180 |
+
|
| 181 |
+
with open(configs_json, "w") as f:
|
| 182 |
+
json.dump(configs, f, indent=4)
|
| 183 |
+
|
| 184 |
+
gr_info(translations["success"])
|
| 185 |
+
return "fp16" if fp16 else "fp32"
|
| 186 |
+
|
| 187 |
+
def process_output(file_path):
|
| 188 |
+
if config.configs.get("delete_exists_file", True):
|
| 189 |
+
if os.path.exists(file_path) and os.path.isfile(file_path): os.remove(file_path)
|
| 190 |
+
return file_path
|
| 191 |
+
else:
|
| 192 |
+
if not os.path.exists(file_path): return file_path
|
| 193 |
+
file = os.path.splitext(os.path.basename(file_path))
|
| 194 |
+
|
| 195 |
+
index = 1
|
| 196 |
+
while 1:
|
| 197 |
+
file_path = os.path.join(os.path.dirname(file_path), f"{file[0]}_{index}{file[1]}")
|
| 198 |
+
if not os.path.exists(file_path): return file_path
|
| 199 |
+
index += 1
|
| 200 |
+
|
| 201 |
+
def shutil_move(input_path, output_path):
|
| 202 |
+
output_path = os.path.join(output_path, os.path.basename(input_path)) if os.path.isdir(output_path) else output_path
|
| 203 |
+
|
| 204 |
+
return shutil.move(input_path, process_output(output_path)) if os.path.exists(output_path) else shutil.move(input_path, output_path)
|
| 205 |
+
|
| 206 |
+
def separate_change(model_name, karaoke_model, reverb_model, enable_post_process, separate_backing, separate_reverb, enable_denoise):
|
| 207 |
+
model_type = "vr" if model_name in list(vr_models.keys()) else "mdx" if model_name in list(mdx_models.keys()) else "demucs" if model_name in list(demucs_models.keys()) else ""
|
| 208 |
+
karaoke_type = ("vr" if karaoke_model.startswith("VR") else "mdx") if separate_backing else None
|
| 209 |
+
reverb_type = ("vr" if not reverb_model.startswith("MDX") else "mdx") if separate_reverb else None
|
| 210 |
+
|
| 211 |
+
all_types = {model_type, karaoke_type, reverb_type}
|
| 212 |
+
|
| 213 |
+
is_vr = "vr" in all_types
|
| 214 |
+
is_mdx = "mdx" in all_types
|
| 215 |
+
is_demucs = "demucs" in all_types
|
| 216 |
+
|
| 217 |
+
return [
|
| 218 |
+
visible(separate_backing),
|
| 219 |
+
visible(separate_reverb),
|
| 220 |
+
visible(is_mdx or is_demucs),
|
| 221 |
+
visible(is_mdx or is_demucs),
|
| 222 |
+
visible(is_mdx),
|
| 223 |
+
visible(is_mdx or is_vr),
|
| 224 |
+
visible(is_demucs),
|
| 225 |
+
visible(is_vr),
|
| 226 |
+
visible(is_vr),
|
| 227 |
+
visible(is_vr and enable_post_process),
|
| 228 |
+
visible(is_vr and enable_denoise),
|
| 229 |
+
valueFalse_interactive(is_vr),
|
| 230 |
+
valueFalse_interactive(is_vr),
|
| 231 |
+
valueFalse_interactive(is_vr)
|
| 232 |
+
]
|
| 233 |
+
|
| 234 |
+
def create_dataset_change(model_name, reverb_model, enable_post_process, separate_reverb, enable_denoise):
|
| 235 |
+
model_type = "vr" if model_name in list(vr_models.keys()) else "mdx" if model_name in list(mdx_models.keys()) else "demucs" if model_name in list(demucs_models.keys()) else ""
|
| 236 |
+
reverb_type = ("vr" if not reverb_model.startswith("MDX") else "mdx") if separate_reverb else None
|
| 237 |
+
all_types = {model_type, reverb_type}
|
| 238 |
+
|
| 239 |
+
is_vr = "vr" in all_types
|
| 240 |
+
is_mdx = "mdx" in all_types
|
| 241 |
+
is_demucs = "demucs" in all_types
|
| 242 |
+
|
| 243 |
+
return [
|
| 244 |
+
visible(separate_reverb),
|
| 245 |
+
visible(is_mdx or is_demucs),
|
| 246 |
+
visible(is_mdx or is_demucs),
|
| 247 |
+
visible(is_mdx),
|
| 248 |
+
visible(is_mdx or is_vr),
|
| 249 |
+
visible(is_demucs),
|
| 250 |
+
visible(is_vr),
|
| 251 |
+
visible(is_vr),
|
| 252 |
+
visible(is_vr and enable_post_process),
|
| 253 |
+
visible(is_vr and enable_denoise),
|
| 254 |
+
valueFalse_interactive(is_vr),
|
| 255 |
+
valueFalse_interactive(is_vr),
|
| 256 |
+
valueFalse_interactive(is_vr)
|
| 257 |
+
]
|
| 258 |
+
|
| 259 |
+
def audio_device():
|
| 260 |
+
try:
|
| 261 |
+
input_devices, output_devices = list_audio_device()
|
| 262 |
+
|
| 263 |
+
def priority(name):
|
| 264 |
+
n = name.lower()
|
| 265 |
+
if "virtual" in n:
|
| 266 |
+
return 0
|
| 267 |
+
if "vb" in n:
|
| 268 |
+
return 1
|
| 269 |
+
return 2
|
| 270 |
+
|
| 271 |
+
output_sorted = sorted(output_devices, key=lambda d: priority(d.name))
|
| 272 |
+
input_sorted = sorted(
|
| 273 |
+
input_devices, key=lambda d: priority(d.name), reverse=True
|
| 274 |
+
)
|
| 275 |
+
|
| 276 |
+
input_device_list = {
|
| 277 |
+
f"{input_sorted.index(d)+1}: {d.name} ({d.host_api})": [d.index, d.max_input_channels] for d in input_sorted
|
| 278 |
+
}
|
| 279 |
+
output_device_list = {
|
| 280 |
+
f"{output_sorted.index(d)+1}: {d.name} ({d.host_api})": [d.index, d.max_output_channels] for d in output_sorted
|
| 281 |
+
}
|
| 282 |
+
|
| 283 |
+
return input_device_list, output_device_list
|
| 284 |
+
except Exception:
|
| 285 |
+
return [], []
|
| 286 |
+
|
| 287 |
+
def update_audio_device(input_device, output_device, monitor_device, monitor):
|
| 288 |
+
input_channels_map, output_channels_map = audio_device()
|
| 289 |
+
|
| 290 |
+
input_is_asio = "ASIO" in input_device if input_device else False
|
| 291 |
+
output_is_asio = "ASIO" in output_device if output_device else False
|
| 292 |
+
monitor_is_asio = "ASIO" in monitor_device if monitor_device else False
|
| 293 |
+
|
| 294 |
+
try:
|
| 295 |
+
input_max_ch = input_channels_map.get(input_device, [])[1]
|
| 296 |
+
output_max_ch = output_channels_map.get(output_device, [])[1]
|
| 297 |
+
monitor_max_ch = output_channels_map.get(monitor_device, [])[1] if monitor else 128
|
| 298 |
+
except:
|
| 299 |
+
input_max_ch = output_max_ch = monitor_max_ch = -1
|
| 300 |
+
|
| 301 |
+
return [
|
| 302 |
+
visible(monitor),
|
| 303 |
+
visible(monitor),
|
| 304 |
+
visible(monitor_is_asio),
|
| 305 |
+
visible(input_is_asio or output_is_asio or monitor_is_asio),
|
| 306 |
+
gr.update(visible=input_is_asio, maximum=input_max_ch),
|
| 307 |
+
gr.update(visible=output_is_asio, maximum=output_max_ch),
|
| 308 |
+
gr.update(visible=monitor_is_asio, maximum=monitor_max_ch)
|
| 309 |
+
]
|
| 310 |
+
|
| 311 |
+
def change_audio_device_choices():
|
| 312 |
+
sd._terminate()
|
| 313 |
+
sd._initialize()
|
| 314 |
+
|
| 315 |
+
input_channels_map, output_channels_map = audio_device()
|
| 316 |
+
input_channels_map, output_channels_map = list(input_channels_map.keys()), list(output_channels_map.keys())
|
| 317 |
+
|
| 318 |
+
return [
|
| 319 |
+
{"value": input_channels_map[0] if len(input_channels_map) >= 1 else "", "choices": input_channels_map, "__type__": "update"},
|
| 320 |
+
{"value": output_channels_map[0] if len(output_channels_map) >= 1 else "", "choices": output_channels_map, "__type__": "update"},
|
| 321 |
+
{"value": output_channels_map[0] if len(output_channels_map) >= 1 else "", "choices": output_channels_map, "__type__": "update"}
|
| 322 |
+
]
|
| 323 |
+
|
| 324 |
+
def replace_punctuation(filename):
|
| 325 |
+
return filename.replace(" ", "_").replace("-", "").replace("(", "").replace(")", "").replace("[", "").replace("]", "").replace(",", "").replace('"', "").replace("'", "").replace("|", "_").replace("{", "").replace("}", "").replace("-_-", "_").replace("_-_", "_").replace("-", "_").replace("---", "_").replace("___", "_").strip()
|
| 326 |
+
|
| 327 |
+
def replace_url(url):
|
| 328 |
+
return url.replace("/blob/", "/resolve/").replace("?download=true", "").strip()
|
| 329 |
+
|
| 330 |
+
def replace_modelname(modelname):
|
| 331 |
+
return replace_punctuation(modelname.replace(".onnx", "").replace(".pth", "").replace(".index", "").replace(".zip", ""))
|
| 332 |
+
|
| 333 |
+
def replace_export_format(audio_path, export_format = "wav"):
|
| 334 |
+
export_format = f".{export_format}"
|
| 335 |
+
|
| 336 |
+
return audio_path if audio_path.endswith(export_format) else audio_path.replace(f".{os.path.basename(audio_path).split('.')[-1]}", export_format)
|
| 337 |
+
|
| 338 |
+
def update_dropdowns_from_json(data):
|
| 339 |
+
if not data:
|
| 340 |
+
return [
|
| 341 |
+
gr.update(choices=[], value=None),
|
| 342 |
+
gr.update(choices=[], value=None),
|
| 343 |
+
gr.update(choices=[], value=None)
|
| 344 |
+
]
|
| 345 |
+
|
| 346 |
+
inputs = list(data.get("inputs", {}).keys())
|
| 347 |
+
outputs = list(data.get("outputs", {}).keys())
|
| 348 |
+
|
| 349 |
+
return [
|
| 350 |
+
gr.update(choices=inputs, value=inputs[0] if len(inputs) > 0 else None),
|
| 351 |
+
gr.update(choices=outputs, value=outputs[0] if len(outputs) > 0 else None),
|
| 352 |
+
gr.update(choices=outputs, value=outputs[0] if len(outputs) > 0 else None),
|
| 353 |
+
]
|
| 354 |
+
|
| 355 |
+
def update_button_from_json(data):
|
| 356 |
+
if not data:
|
| 357 |
+
return [gr.update(interactive=True), gr.update(interactive=False)]
|
| 358 |
+
|
| 359 |
+
return [
|
| 360 |
+
gr.update(interactive=data.get("start_button", True)),
|
| 361 |
+
gr.update(interactive=data.get("stop_button", False))
|
| 362 |
]
|