Spaces:
Build error
Build error
File size: 17,009 Bytes
f1f433f |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 |
import os
import platform
from ctypes import CDLL, POINTER, c_bool, c_char_p, c_float, c_int, c_long
from ctypes.util import find_library
from dataclasses import dataclass
from enum import Enum, auto
from pathlib import Path
from typing import List, Optional
import numpy as np
class OldCoreError(Exception):
"""古いコアが使用されている場合に発生するエラー"""
class CoreError(Exception):
"""コア呼び出しで発生したエラー"""
def load_runtime_lib(runtime_dirs: List[Path]):
if platform.system() == "Windows":
# DirectML.dllはonnxruntimeと互換性のないWindows標準搭載のものを優先して読み込むことがあるため、明示的に読み込む
# 参考 1. https://github.com/microsoft/onnxruntime/issues/3360
# 参考 2. https://tadaoyamaoka.hatenablog.com/entry/2020/06/07/113616
lib_file_names = [
"torch_cpu.dll",
"torch_cuda.dll",
"DirectML.dll",
"onnxruntime.dll",
]
lib_names = ["torch_cpu", "torch_cuda", "onnxruntime"]
elif platform.system() == "Linux":
lib_file_names = ["libtorch.so", "libonnxruntime.so"]
lib_names = ["torch", "onnxruntime"]
elif platform.system() == "Darwin":
lib_file_names = ["libonnxruntime.dylib"]
lib_names = ["onnxruntime"]
else:
raise RuntimeError("不明なOSです")
for lib_path in runtime_dirs:
for file_name in lib_file_names:
try:
CDLL(str((lib_path / file_name).resolve(strict=True)))
except OSError:
pass
for lib_name in lib_names:
try:
CDLL(find_library(lib_name))
except (OSError, TypeError):
pass
class GPUType(Enum):
# NONEはCPUしか対応していないことを示す
NONE = auto()
CUDA = auto()
DIRECT_ML = auto()
@dataclass(frozen=True)
class CoreInfo:
name: str
platform: str
arch: str
core_type: str
gpu_type: GPUType
# version 0.12 より前のコアの情報
CORE_INFOS = [
# Windows
CoreInfo(
name="core.dll",
platform="Windows",
arch="x64",
core_type="libtorch",
gpu_type=GPUType.CUDA,
),
CoreInfo(
name="core_cpu.dll",
platform="Windows",
arch="x64",
core_type="libtorch",
gpu_type=GPUType.NONE,
),
CoreInfo(
name="core_gpu_x64_nvidia.dll",
platform="Windows",
arch="x64",
core_type="onnxruntime",
gpu_type=GPUType.CUDA,
),
CoreInfo(
name="core_gpu_x64_directml.dll",
platform="Windows",
arch="x64",
core_type="onnxruntime",
gpu_type=GPUType.DIRECT_ML,
),
CoreInfo(
name="core_cpu_x64.dll",
platform="Windows",
arch="x64",
core_type="onnxruntime",
gpu_type=GPUType.NONE,
),
CoreInfo(
name="core_cpu_x86.dll",
platform="Windows",
arch="x86",
core_type="onnxruntime",
gpu_type=GPUType.NONE,
),
CoreInfo(
name="core_gpu_x86_directml.dll",
platform="Windows",
arch="x86",
core_type="onnxruntime",
gpu_type=GPUType.DIRECT_ML,
),
CoreInfo(
name="core_cpu_arm.dll",
platform="Windows",
arch="armv7l",
core_type="onnxruntime",
gpu_type=GPUType.NONE,
),
CoreInfo(
name="core_gpu_arm_directml.dll",
platform="Windows",
arch="armv7l",
core_type="onnxruntime",
gpu_type=GPUType.DIRECT_ML,
),
CoreInfo(
name="core_cpu_arm64.dll",
platform="Windows",
arch="aarch64",
core_type="onnxruntime",
gpu_type=GPUType.NONE,
),
CoreInfo(
name="core_gpu_arm64_directml.dll",
platform="Windows",
arch="aarch64",
core_type="onnxruntime",
gpu_type=GPUType.DIRECT_ML,
),
# Linux
CoreInfo(
name="libcore.so",
platform="Linux",
arch="x64",
core_type="libtorch",
gpu_type=GPUType.CUDA,
),
CoreInfo(
name="libcore_cpu.so",
platform="Linux",
arch="x64",
core_type="libtorch",
gpu_type=GPUType.NONE,
),
CoreInfo(
name="libcore_gpu_x64_nvidia.so",
platform="Linux",
arch="x64",
core_type="onnxruntime",
gpu_type=GPUType.CUDA,
),
CoreInfo(
name="libcore_cpu_x64.so",
platform="Linux",
arch="x64",
core_type="onnxruntime",
gpu_type=GPUType.NONE,
),
CoreInfo(
name="libcore_cpu_armhf.so",
platform="Linux",
arch="armv7l",
core_type="onnxruntime",
gpu_type=GPUType.NONE,
),
CoreInfo(
name="libcore_cpu_arm64.so",
platform="Linux",
arch="aarch64",
core_type="onnxruntime",
gpu_type=GPUType.NONE,
),
# macOS
CoreInfo(
name="libcore_cpu_universal2.dylib",
platform="Darwin",
arch="universal",
core_type="onnxruntime",
gpu_type=GPUType.NONE,
),
]
# version 0.12 以降のコアの名前の辞書
# - version 0.12, 0.13 のコアの名前: core
# - version 0.14 からのコアの名前: voicevox_core
CORENAME_DICT = {
"Windows": ("voicevox_core.dll", "core.dll"),
"Linux": ("libvoicevox_core.so", "libcore.so"),
"Darwin": ("libvoicevox_core.dylib", "libcore.dylib"),
}
def find_version_0_12_core_or_later(core_dir: Path) -> Optional[str]:
"""
core_dir で指定したディレクトリにあるコアライブラリが Version 0.12 以降である場合、
見つかった共有ライブラリの名前を返す。
Version 0.12 以降と判定する条件は、
- core_dir に metas.json が存在しない
- コアライブラリの名前が CORENAME_DICT の定義に従っている
の両方が真のときである。
cf. https://github.com/VOICEVOX/voicevox_engine/issues/385
"""
if (core_dir / "metas.json").exists():
return None
for core_name in CORENAME_DICT[platform.system()]:
if (core_dir / core_name).is_file():
return core_name
return None
def get_arch_name() -> Optional[str]:
"""
platform.machine() が特定のアーキテクチャ上で複数パターンの文字列を返し得るので、
一意な文字列に変換する
サポート外のアーキテクチャである場合、None を返す
"""
machine = platform.machine()
if machine == "x86_64" or machine == "x64" or machine == "AMD64":
return "x64"
elif machine == "i386" or machine == "x86":
return "x86"
elif machine == "arm64":
return "aarch64"
elif machine in ["armv7l", "aarch64"]:
return machine
else:
return None
def get_core_name(
arch_name: str,
platform_name: str,
model_type: str,
gpu_type: GPUType,
) -> Optional[str]:
if platform_name == "Darwin":
if gpu_type == GPUType.NONE and (arch_name == "x64" or arch_name == "aarch64"):
arch_name = "universal"
else:
return None
for core_info in CORE_INFOS:
if (
core_info.platform == platform_name
and core_info.arch == arch_name
and core_info.core_type == model_type
and core_info.gpu_type == gpu_type
):
return core_info.name
return None
def get_suitable_core_name(
model_type: str,
gpu_type: GPUType,
) -> Optional[str]:
arch_name = get_arch_name()
if arch_name is None:
return None
platform_name = platform.system()
return get_core_name(arch_name, platform_name, model_type, gpu_type)
def check_core_type(core_dir: Path) -> Optional[str]:
# libtorch版はDirectML未対応なので、ここでは`gpu_type=GPUType.DIRECT_ML`は入れない
libtorch_core_names = [
get_suitable_core_name("libtorch", gpu_type=GPUType.CUDA),
get_suitable_core_name("libtorch", gpu_type=GPUType.NONE),
]
onnxruntime_core_names = [
get_suitable_core_name("onnxruntime", gpu_type=GPUType.CUDA),
get_suitable_core_name("onnxruntime", gpu_type=GPUType.DIRECT_ML),
get_suitable_core_name("onnxruntime", gpu_type=GPUType.NONE),
]
if any([(core_dir / name).is_file() for name in libtorch_core_names if name]):
return "libtorch"
elif any([(core_dir / name).is_file() for name in onnxruntime_core_names if name]):
return "onnxruntime"
else:
return None
def load_core(core_dir: Path, use_gpu: bool) -> CDLL:
core_name = find_version_0_12_core_or_later(core_dir)
if core_name:
try:
# NOTE: CDLL クラスのコンストラクタの引数 name には文字列を渡す必要がある。
# Windows 環境では PathLike オブジェクトを引数として渡すと初期化に失敗する。
return CDLL(str((core_dir / core_name).resolve(strict=True)))
except OSError as err:
raise RuntimeError(f"コアの読み込みに失敗しました:{err}")
model_type = check_core_type(core_dir)
if model_type is None:
raise RuntimeError("コアが見つかりません")
if use_gpu or model_type == "onnxruntime":
core_name = get_suitable_core_name(model_type, gpu_type=GPUType.CUDA)
if core_name:
try:
return CDLL(str((core_dir / core_name).resolve(strict=True)))
except OSError:
pass
core_name = get_suitable_core_name(model_type, gpu_type=GPUType.DIRECT_ML)
if core_name:
try:
return CDLL(str((core_dir / core_name).resolve(strict=True)))
except OSError:
pass
core_name = get_suitable_core_name(model_type, gpu_type=GPUType.NONE)
if core_name:
try:
return CDLL(str((core_dir / core_name).resolve(strict=True)))
except OSError as err:
if model_type == "libtorch":
core_name = get_suitable_core_name(model_type, gpu_type=GPUType.CUDA)
if core_name:
try:
return CDLL(str((core_dir / core_name).resolve(strict=True)))
except OSError as err_:
err = err_
raise RuntimeError(f"コアの読み込みに失敗しました:{err}")
else:
raise RuntimeError(f"このコンピュータのアーキテクチャ {platform.machine()} で利用可能なコアがありません")
class CoreWrapper:
def __init__(
self,
use_gpu: bool,
core_dir: Path,
cpu_num_threads: int = 0,
load_all_models: bool = False,
) -> None:
self.core = load_core(core_dir, use_gpu)
self.core.initialize.restype = c_bool
self.core.metas.restype = c_char_p
self.core.yukarin_s_forward.restype = c_bool
self.core.yukarin_sa_forward.restype = c_bool
self.core.decode_forward.restype = c_bool
self.core.last_error_message.restype = c_char_p
self.exist_supported_devices = False
self.exist_finalize = False
exist_cpu_num_threads = False
self.exist_load_model = False
self.exist_is_model_loaded = False
is_version_0_12_core_or_later = (
find_version_0_12_core_or_later(core_dir) is not None
)
if is_version_0_12_core_or_later:
model_type = "onnxruntime"
self.exist_load_model = True
self.exist_is_model_loaded = True
self.core.load_model.argtypes = (c_long,)
self.core.load_model.restype = c_bool
self.core.is_model_loaded.argtypes = (c_long,)
self.core.is_model_loaded.restype = c_bool
else:
model_type = check_core_type(core_dir)
assert model_type is not None
if model_type == "onnxruntime":
self.core.supported_devices.restype = c_char_p
self.core.finalize.restype = None
self.exist_supported_devices = True
self.exist_finalize = True
exist_cpu_num_threads = True
self.core.yukarin_s_forward.argtypes = (
c_int,
POINTER(c_long),
POINTER(c_long),
POINTER(c_float),
)
self.core.yukarin_sa_forward.argtypes = (
c_int,
POINTER(c_long),
POINTER(c_long),
POINTER(c_long),
POINTER(c_long),
POINTER(c_long),
POINTER(c_long),
POINTER(c_long),
POINTER(c_float),
)
self.core.decode_forward.argtypes = (
c_int,
c_int,
POINTER(c_float),
POINTER(c_float),
POINTER(c_long),
POINTER(c_float),
)
cwd = os.getcwd()
os.chdir(core_dir)
try:
if is_version_0_12_core_or_later:
self.assert_core_success(
self.core.initialize(use_gpu, cpu_num_threads, load_all_models)
)
elif exist_cpu_num_threads:
self.assert_core_success(
self.core.initialize(".", use_gpu, cpu_num_threads)
)
else:
self.assert_core_success(self.core.initialize(".", use_gpu))
finally:
os.chdir(cwd)
def metas(self) -> str:
return self.core.metas().decode("utf-8")
def yukarin_s_forward(
self,
length: int,
phoneme_list: np.ndarray,
speaker_id: np.ndarray,
) -> np.ndarray:
output = np.zeros((length,), dtype=np.float32)
self.assert_core_success(
self.core.yukarin_s_forward(
c_int(length),
phoneme_list.ctypes.data_as(POINTER(c_long)),
speaker_id.ctypes.data_as(POINTER(c_long)),
output.ctypes.data_as(POINTER(c_float)),
)
)
return output
def yukarin_sa_forward(
self,
length: int,
vowel_phoneme_list: np.ndarray,
consonant_phoneme_list: np.ndarray,
start_accent_list: np.ndarray,
end_accent_list: np.ndarray,
start_accent_phrase_list: np.ndarray,
end_accent_phrase_list: np.ndarray,
speaker_id: np.ndarray,
) -> np.ndarray:
output = np.empty(
(
len(speaker_id),
length,
),
dtype=np.float32,
)
self.assert_core_success(
self.core.yukarin_sa_forward(
c_int(length),
vowel_phoneme_list.ctypes.data_as(POINTER(c_long)),
consonant_phoneme_list.ctypes.data_as(POINTER(c_long)),
start_accent_list.ctypes.data_as(POINTER(c_long)),
end_accent_list.ctypes.data_as(POINTER(c_long)),
start_accent_phrase_list.ctypes.data_as(POINTER(c_long)),
end_accent_phrase_list.ctypes.data_as(POINTER(c_long)),
speaker_id.ctypes.data_as(POINTER(c_long)),
output.ctypes.data_as(POINTER(c_float)),
)
)
return output
def decode_forward(
self,
length: int,
phoneme_size: int,
f0: np.ndarray,
phoneme: np.ndarray,
speaker_id: np.ndarray,
) -> np.ndarray:
output = np.empty((length * 256,), dtype=np.float32)
self.assert_core_success(
self.core.decode_forward(
c_int(length),
c_int(phoneme_size),
f0.ctypes.data_as(POINTER(c_float)),
phoneme.ctypes.data_as(POINTER(c_float)),
speaker_id.ctypes.data_as(POINTER(c_long)),
output.ctypes.data_as(POINTER(c_float)),
)
)
return output
def supported_devices(self) -> str:
if self.exist_supported_devices:
return self.core.supported_devices().decode("utf-8")
raise OldCoreError
def finalize(self) -> None:
if self.exist_finalize:
self.core.finalize()
return
raise OldCoreError
def load_model(self, speaker_id: int) -> None:
if self.exist_load_model:
self.assert_core_success(self.core.load_model(c_long(speaker_id)))
raise OldCoreError
def is_model_loaded(self, speaker_id: int) -> bool:
if self.exist_is_model_loaded:
return self.core.is_model_loaded(c_long(speaker_id))
raise OldCoreError
def assert_core_success(self, result: bool) -> None:
if not result:
raise CoreError(
self.core.last_error_message().decode("utf-8", "backslashreplace")
)
|