zh-xiaoya-1p55m/manifest.json
| 1 | { |
| 2 | "format": "roota.raw-fp16.v1", |
| 3 | "package_name": "zh_CN", |
| 4 | "language": "zh_CN", |
| 5 | "voice": "zh_CN-xiao_ya-medium", |
| 6 | "created_from": "tools/export_roota_self_contained_package.py", |
| 7 | "sample_rate": 22050, |
| 8 | "hop_length": 256, |
| 9 | "inference": { |
| 10 | "duration_length_scale": 1.0, |
| 11 | "duration_policy": "round(exp(log_duration) * duration_length_scale)" |
| 12 | }, |
| 13 | "weights_file": "weights.fp16.bin", |
| 14 | "weights_size_bytes": 3108568, |
| 15 | "weights_sha256": "ec4faabed7e443a3eab3517f8d1d353624babaadd8357213e83def82507f7497", |
| 16 | "total_parameters": 1554284, |
| 17 | "components": { |
| 18 | "duration": { |
| 19 | "checkpoint": "artifacts/voices/zh_CN/duration/duration-student.pt", |
| 20 | "checkpoint_sha256": "2db8ed13a01eb7638f056aca2edcd576eedc13683745307715f1e5d2c6dcbb00", |
| 21 | "config": { |
| 22 | "architecture": "duration_conv", |
| 23 | "vocab_size": 73, |
| 24 | "hidden": 64, |
| 25 | "depth": 3, |
| 26 | "kernel_size": 5, |
| 27 | "max_tokens": 432, |
| 28 | "max_duration": 80, |
| 29 | "target_duration_scale": 1.0, |
| 30 | "pause_preserve_scale": 0.0, |
| 31 | "pause_preserve_window": 0, |
| 32 | "pause_token_ids": [ |
| 33 | 4, |
| 34 | 8, |
| 35 | 10, |
| 36 | 11, |
| 37 | 12, |
| 38 | 13 |
| 39 | ], |
| 40 | "long_preserve_threshold": 0.0, |
| 41 | "long_preserve_scale": 0.0, |
| 42 | "duration_vocab_size_forced": 73, |
| 43 | "duration_oov_id": 0 |
| 44 | }, |
| 45 | "parameters": 132356, |
| 46 | "tensors": [ |
| 47 | { |
| 48 | "name": "blocks.0.net.0.bias", |
| 49 | "shape": [ |
| 50 | 64 |
| 51 | ], |
| 52 | "dtype": "float16", |
| 53 | "offset_bytes": 0, |
| 54 | "nbytes": 128, |
| 55 | "sha256": "eaa982bd16fa987e43de8d19588e23054c939c01bd78ba84254286d25c489c56" |
| 56 | }, |
| 57 | { |
| 58 | "name": "blocks.0.net.0.weight", |
| 59 | "shape": [ |
| 60 | 64, |
| 61 | 64, |
| 62 | 5 |
| 63 | ], |
| 64 | "dtype": "float16", |
| 65 | "offset_bytes": 128, |
| 66 | "nbytes": 40960, |
| 67 | "sha256": "f75c79be7a78edd2a4edcdd6195960a0c96a5e00dff31adfdeac5c9090e04985" |
| 68 | }, |
| 69 | { |
| 70 | "name": "blocks.0.net.2.bias", |
| 71 | "shape": [ |
| 72 | 64 |
| 73 | ], |
| 74 | "dtype": "float16", |
| 75 | "offset_bytes": 41088, |
| 76 | "nbytes": 128, |
| 77 | "sha256": "1ba18ea1142d3757599ed8f7b260f57ef9cf1240e836677aafbef4354c2d7be4" |
| 78 | }, |
| 79 | { |
| 80 | "name": "blocks.0.net.2.weight", |
| 81 | "shape": [ |
| 82 | 64, |
| 83 | 64, |
| 84 | 5 |
| 85 | ], |
| 86 | "dtype": "float16", |
| 87 | "offset_bytes": 41216, |
| 88 | "nbytes": 40960, |
| 89 | "sha256": "e7d61de8ac78c410bc8e7100c9eba916351b52b3c702f5951cc05029fa464993" |
| 90 | }, |
| 91 | { |
| 92 | "name": "blocks.0.scale", |
| 93 | "shape": [ |
| 94 | 1 |
| 95 | ], |
| 96 | "dtype": "float16", |
| 97 | "offset_bytes": 82176, |
| 98 | "nbytes": 2, |
| 99 | "sha256": "aea77e50e348d9cbd033698b23754d314778a6db55f3b65db1bc0f0063c47090" |
| 100 | }, |
| 101 | { |
| 102 | "name": "blocks.1.net.0.bias", |
| 103 | "shape": [ |
| 104 | 64 |
| 105 | ], |
| 106 | "dtype": "float16", |
| 107 | "offset_bytes": 82178, |
| 108 | "nbytes": 128, |
| 109 | "sha256": "40fe6a6dacfcaab5ea2e56a65c93411851312286df4bb9be61f4c8fa3a12dd52" |
| 110 | }, |
| 111 | { |
| 112 | "name": "blocks.1.net.0.weight", |
| 113 | "shape": [ |
| 114 | 64, |
| 115 | 64, |
| 116 | 5 |
| 117 | ], |
| 118 | "dtype": "float16", |
| 119 | "offset_bytes": 82306, |
| 120 | "nbytes": 40960, |
| 121 | "sha256": "bfb9ecd5d67ff1d8b1ba8c14e4dc938b4ad9bed15ae559237e1469a289993111" |
| 122 | }, |
| 123 | { |
| 124 | "name": "blocks.1.net.2.bias", |
| 125 | "shape": [ |
| 126 | 64 |
| 127 | ], |
| 128 | "dtype": "float16", |
| 129 | "offset_bytes": 123266, |
| 130 | "nbytes": 128, |
| 131 | "sha256": "88ba26cbce98cd87f09d2e8fdf79ce9d91ad19d8fe44ae1b076fd80a34f201a7" |
| 132 | }, |
| 133 | { |
| 134 | "name": "blocks.1.net.2.weight", |
| 135 | "shape": [ |
| 136 | 64, |
| 137 | 64, |
| 138 | 5 |
| 139 | ], |
| 140 | "dtype": "float16", |
| 141 | "offset_bytes": 123394, |
| 142 | "nbytes": 40960, |
| 143 | "sha256": "0f932882d56b8fc234e68f44078673625161c3f02cde2550a9a5c00fb73d5c86" |
| 144 | }, |
| 145 | { |
| 146 | "name": "blocks.1.scale", |
| 147 | "shape": [ |
| 148 | 1 |
| 149 | ], |
| 150 | "dtype": "float16", |
| 151 | "offset_bytes": 164354, |
| 152 | "nbytes": 2, |
| 153 | "sha256": "c1c7c8f2e597fc4c2ef95cb505476a4b301edfad69c03f79d6baf338650112ea" |
| 154 | }, |
| 155 | { |
| 156 | "name": "blocks.2.net.0.bias", |
| 157 | "shape": [ |
| 158 | 64 |
| 159 | ], |
| 160 | "dtype": "float16", |
| 161 | "offset_bytes": 164356, |
| 162 | "nbytes": 128, |
| 163 | "sha256": "0b1ba2336ef8c3411213e4874e425ecbc403dce710dc696e58e51db6135b7b16" |
| 164 | }, |
| 165 | { |
| 166 | "name": "blocks.2.net.0.weight", |
| 167 | "shape": [ |
| 168 | 64, |
| 169 | 64, |
| 170 | 5 |
| 171 | ], |
| 172 | "dtype": "float16", |
| 173 | "offset_bytes": 164484, |
| 174 | "nbytes": 40960, |
| 175 | "sha256": "0b07fc1ea7c3de187f3610dee72e2f122c4b6cc617e9b0f59189035ac86c8591" |
| 176 | }, |
| 177 | { |
| 178 | "name": "blocks.2.net.2.bias", |
| 179 | "shape": [ |
| 180 | 64 |
| 181 | ], |
| 182 | "dtype": "float16", |
| 183 | "offset_bytes": 205444, |
| 184 | "nbytes": 128, |
| 185 | "sha256": "1ad18aa3c70166b70023980e74771274d4119169de93f285155dfda6d2114773" |
| 186 | }, |
| 187 | { |
| 188 | "name": "blocks.2.net.2.weight", |
| 189 | "shape": [ |
| 190 | 64, |
| 191 | 64, |
| 192 | 5 |
| 193 | ], |
| 194 | "dtype": "float16", |
| 195 | "offset_bytes": 205572, |
| 196 | "nbytes": 40960, |
| 197 | "sha256": "75e3e122b67c26a22227c63fd108ba4291d528f681dde390ad819adc00a17791" |
| 198 | }, |
| 199 | { |
| 200 | "name": "blocks.2.scale", |
| 201 | "shape": [ |
| 202 | 1 |
| 203 | ], |
| 204 | "dtype": "float16", |
| 205 | "offset_bytes": 246532, |
| 206 | "nbytes": 2, |
| 207 | "sha256": "0d96c0c12f2ccf5f607e4c3bea40f3c0b8e8e1eac20a4caa192c3645141002b3" |
| 208 | }, |
| 209 | { |
| 210 | "name": "embedding.weight", |
| 211 | "shape": [ |
| 212 | 73, |
| 213 | 64 |
| 214 | ], |
| 215 | "dtype": "float16", |
| 216 | "offset_bytes": 246534, |
| 217 | "nbytes": 9344, |
| 218 | "sha256": "abac44b3ad2d0024dee542f809f6e5085b9681a3381b57d1f7d3f702d7e51795" |
| 219 | }, |
| 220 | { |
| 221 | "name": "input_proj.bias", |
| 222 | "shape": [ |
| 223 | 64 |
| 224 | ], |
| 225 | "dtype": "float16", |
| 226 | "offset_bytes": 255878, |
| 227 | "nbytes": 128, |
| 228 | "sha256": "70bf98bc75f715577f9857d8a18872b6cc465f15c9a9d1861e69da1c421bafae" |
| 229 | }, |
| 230 | { |
| 231 | "name": "input_proj.weight", |
| 232 | "shape": [ |
| 233 | 64, |
| 234 | 67, |
| 235 | 1 |
| 236 | ], |
| 237 | "dtype": "float16", |
| 238 | "offset_bytes": 256006, |
| 239 | "nbytes": 8576, |
| 240 | "sha256": "daba0b9940f15004b4ed3e181bbc1c6ffa18471a2799800fe8777263ad06d410" |
| 241 | }, |
| 242 | { |
| 243 | "name": "output.bias", |
| 244 | "shape": [ |
| 245 | 1 |
| 246 | ], |
| 247 | "dtype": "float16", |
| 248 | "offset_bytes": 264582, |
| 249 | "nbytes": 2, |
| 250 | "sha256": "c02c0b965e023abee808f2b548d8d5193a8b5229be6f3121a6f16e2d41a449b3" |
| 251 | }, |
| 252 | { |
| 253 | "name": "output.weight", |
| 254 | "shape": [ |
| 255 | 1, |
| 256 | 64, |
| 257 | 1 |
| 258 | ], |
| 259 | "dtype": "float16", |
| 260 | "offset_bytes": 264584, |
| 261 | "nbytes": 128, |
| 262 | "sha256": "3f50093f8f88cc1bd00d7e51bed01298bf301d591033f844d947c839b02994c2" |
| 263 | } |
| 264 | ] |
| 265 | }, |
| 266 | "acoustic": { |
| 267 | "checkpoint": "artifacts/voices/zh_CN/joint/latent-student.pt", |
| 268 | "checkpoint_sha256": "d98261e8ee80eb619659dc0b91374847f4871c8c94f33d3b68abc39da33dd03a", |
| 269 | "config": { |
| 270 | "architecture": "token_context", |
| 271 | "vocab_size": 73, |
| 272 | "hidden": 96, |
| 273 | "depth": 4, |
| 274 | "kernel_size": 5, |
| 275 | "out_channels": 192, |
| 276 | "token_depth": 3 |
| 277 | }, |
| 278 | "parameters": 691207, |
| 279 | "tensors": [ |
| 280 | { |
| 281 | "name": "embedding.weight", |
| 282 | "shape": [ |
| 283 | 73, |
| 284 | 96 |
| 285 | ], |
| 286 | "dtype": "float16", |
| 287 | "offset_bytes": 264712, |
| 288 | "nbytes": 14016, |
| 289 | "sha256": "4b044149ef6910a38658da90148cc3afd8f5af30a42732932f019f48739928db" |
| 290 | }, |
| 291 | { |
| 292 | "name": "frame_blocks.0.net.0.bias", |
| 293 | "shape": [ |
| 294 | 96 |
| 295 | ], |
| 296 | "dtype": "float16", |
| 297 | "offset_bytes": 278728, |
| 298 | "nbytes": 192, |
| 299 | "sha256": "3a49ff82ce1aa9f64517f2bfd609fbbfa108437a58fb9188e3f478830dc90db1" |
| 300 | }, |
| 301 | { |
| 302 | "name": "frame_blocks.0.net.0.weight", |
| 303 | "shape": [ |
| 304 | 96, |
| 305 | 96, |
| 306 | 5 |
| 307 | ], |
| 308 | "dtype": "float16", |
| 309 | "offset_bytes": 278920, |
| 310 | "nbytes": 92160, |
| 311 | "sha256": "d427760b147e6f6183004d79b9bbbf7d05800a85a28ed7cdd6c11180aa7fce7d" |
| 312 | }, |
| 313 | { |
| 314 | "name": "frame_blocks.0.net.2.bias", |
| 315 | "shape": [ |
| 316 | 96 |
| 317 | ], |
| 318 | "dtype": "float16", |
| 319 | "offset_bytes": 371080, |
| 320 | "nbytes": 192, |
| 321 | "sha256": "b266894dfe4f63c834ccbe28cef26378b35d24aa250dbcca052b4ff8334da6d1" |
| 322 | }, |
| 323 | { |
| 324 | "name": "frame_blocks.0.net.2.weight", |
| 325 | "shape": [ |
| 326 | 96, |
| 327 | 96, |
| 328 | 5 |
| 329 | ], |
| 330 | "dtype": "float16", |
| 331 | "offset_bytes": 371272, |
| 332 | "nbytes": 92160, |
| 333 | "sha256": "50eb7900f26afa35896c1eb1b088efc20c8a9d6f09e250ed003068b8467f9c9a" |
| 334 | }, |
| 335 | { |
| 336 | "name": "frame_blocks.0.scale", |
| 337 | "shape": [ |
| 338 | 1 |
| 339 | ], |
| 340 | "dtype": "float16", |
| 341 | "offset_bytes": 463432, |
| 342 | "nbytes": 2, |
| 343 | "sha256": "9f033fe7972d4ac6cde6e04e216cc7161434b240fdc0f8feaeef9cb207cd7f8e" |
| 344 | }, |
| 345 | { |
| 346 | "name": "frame_blocks.1.net.0.bias", |
| 347 | "shape": [ |
| 348 | 96 |
| 349 | ], |
| 350 | "dtype": "float16", |
| 351 | "offset_bytes": 463434, |
| 352 | "nbytes": 192, |
| 353 | "sha256": "092088a7b6d9de013156726048e632e715953377251e69888fcdf639394f63fd" |
| 354 | }, |
| 355 | { |
| 356 | "name": "frame_blocks.1.net.0.weight", |
| 357 | "shape": [ |
| 358 | 96, |
| 359 | 96, |
| 360 | 5 |
| 361 | ], |
| 362 | "dtype": "float16", |
| 363 | "offset_bytes": 463626, |
| 364 | "nbytes": 92160, |
| 365 | "sha256": "7a4bcd51e28ff969649239213c2d340fa9c54dad5fae654ef2d9df5d97e01735" |
| 366 | }, |
| 367 | { |
| 368 | "name": "frame_blocks.1.net.2.bias", |
| 369 | "shape": [ |
| 370 | 96 |
| 371 | ], |
| 372 | "dtype": "float16", |
| 373 | "offset_bytes": 555786, |
| 374 | "nbytes": 192, |
| 375 | "sha256": "b285344d16993f85dfdb5d21008ab0f714dbc6f3777cd1820006f038b9c17ce2" |
| 376 | }, |
| 377 | { |
| 378 | "name": "frame_blocks.1.net.2.weight", |
| 379 | "shape": [ |
| 380 | 96, |
| 381 | 96, |
| 382 | 5 |
| 383 | ], |
| 384 | "dtype": "float16", |
| 385 | "offset_bytes": 555978, |
| 386 | "nbytes": 92160, |
| 387 | "sha256": "983e9f80a244e502b163cb9f0321d9939c2d834043df931a2934fb40bc9af104" |
| 388 | }, |
| 389 | { |
| 390 | "name": "frame_blocks.1.scale", |
| 391 | "shape": [ |
| 392 | 1 |
| 393 | ], |
| 394 | "dtype": "float16", |
| 395 | "offset_bytes": 648138, |
| 396 | "nbytes": 2, |
| 397 | "sha256": "599ee041682c647a30426296462ddea99c2658d94c927c78adf8937d3221a94e" |
| 398 | }, |
| 399 | { |
| 400 | "name": "frame_blocks.2.net.0.bias", |
| 401 | "shape": [ |
| 402 | 96 |
| 403 | ], |
| 404 | "dtype": "float16", |
| 405 | "offset_bytes": 648140, |
| 406 | "nbytes": 192, |
| 407 | "sha256": "53c806182ed10d3647b0614f98968949e907c6c529ce583f01709b85106cd99d" |
| 408 | }, |
| 409 | { |
| 410 | "name": "frame_blocks.2.net.0.weight", |
| 411 | "shape": [ |
| 412 | 96, |
| 413 | 96, |
| 414 | 5 |
| 415 | ], |
| 416 | "dtype": "float16", |
| 417 | "offset_bytes": 648332, |
| 418 | "nbytes": 92160, |
| 419 | "sha256": "9e14a5c75f9d5e59e8492daa17dd61ab053ed984e9fefc33d76285aa62f79408" |
| 420 | }, |
| 421 | { |
| 422 | "name": "frame_blocks.2.net.2.bias", |
| 423 | "shape": [ |
| 424 | 96 |
| 425 | ], |
| 426 | "dtype": "float16", |
| 427 | "offset_bytes": 740492, |
| 428 | "nbytes": 192, |
| 429 | "sha256": "229a26d89fbb32918765441fea214cf9a149cd9795564bd1285606b9643ede77" |
| 430 | }, |
| 431 | { |
| 432 | "name": "frame_blocks.2.net.2.weight", |
| 433 | "shape": [ |
| 434 | 96, |
| 435 | 96, |
| 436 | 5 |
| 437 | ], |
| 438 | "dtype": "float16", |
| 439 | "offset_bytes": 740684, |
| 440 | "nbytes": 92160, |
| 441 | "sha256": "94822e898cdd17fa9c22ddacfea3eed1fa83b79aee8cdb0387f434a976716f15" |
| 442 | }, |
| 443 | { |
| 444 | "name": "frame_blocks.2.scale", |
| 445 | "shape": [ |
| 446 | 1 |
| 447 | ], |
| 448 | "dtype": "float16", |
| 449 | "offset_bytes": 832844, |
| 450 | "nbytes": 2, |
| 451 | "sha256": "4b2d90d434febdd80a53f5acbbdaa5c7bba585d4df757106984afd4e975e6f6d" |
| 452 | }, |
| 453 | { |
| 454 | "name": "frame_blocks.3.net.0.bias", |
| 455 | "shape": [ |
| 456 | 96 |
| 457 | ], |
| 458 | "dtype": "float16", |
| 459 | "offset_bytes": 832846, |
| 460 | "nbytes": 192, |
| 461 | "sha256": "3c8dfa28c692d0d56c641b9f45d9c4dcf0ac6fdc0921052cbd15779d0269cb18" |
| 462 | }, |
| 463 | { |
| 464 | "name": "frame_blocks.3.net.0.weight", |
| 465 | "shape": [ |
| 466 | 96, |
| 467 | 96, |
| 468 | 5 |
| 469 | ], |
| 470 | "dtype": "float16", |
| 471 | "offset_bytes": 833038, |
| 472 | "nbytes": 92160, |
| 473 | "sha256": "d704ec52a04818f4ab9e0fed343604554c08e0cb76342868db127c2a6988a464" |
| 474 | }, |
| 475 | { |
| 476 | "name": "frame_blocks.3.net.2.bias", |
| 477 | "shape": [ |
| 478 | 96 |
| 479 | ], |
| 480 | "dtype": "float16", |
| 481 | "offset_bytes": 925198, |
| 482 | "nbytes": 192, |
| 483 | "sha256": "4623455a5e170bdee6a92a565af668f1e3f1656e55de16ecb03adcb699a4e478" |
| 484 | }, |
| 485 | { |
| 486 | "name": "frame_blocks.3.net.2.weight", |
| 487 | "shape": [ |
| 488 | 96, |
| 489 | 96, |
| 490 | 5 |
| 491 | ], |
| 492 | "dtype": "float16", |
| 493 | "offset_bytes": 925390, |
| 494 | "nbytes": 92160, |
| 495 | "sha256": "3c2d4f1b049396ab9b75076a68dec0f146f8db8559da877ac747dcab292d968f" |
| 496 | }, |
| 497 | { |
| 498 | "name": "frame_blocks.3.scale", |
| 499 | "shape": [ |
| 500 | 1 |
| 501 | ], |
| 502 | "dtype": "float16", |
| 503 | "offset_bytes": 1017550, |
| 504 | "nbytes": 2, |
| 505 | "sha256": "5e90140de066fe5f9ac86ca066e35d648c3f56f13267f54dda956b8f743404e5" |
| 506 | }, |
| 507 | { |
| 508 | "name": "frame_input_proj.bias", |
| 509 | "shape": [ |
| 510 | 96 |
| 511 | ], |
| 512 | "dtype": "float16", |
| 513 | "offset_bytes": 1017552, |
| 514 | "nbytes": 192, |
| 515 | "sha256": "28861954dee53d6043399624d30721ca9fe9b85bf147a6d1de3bf51118436ad0" |
| 516 | }, |
| 517 | { |
| 518 | "name": "frame_input_proj.weight", |
| 519 | "shape": [ |
| 520 | 96, |
| 521 | 99, |
| 522 | 1 |
| 523 | ], |
| 524 | "dtype": "float16", |
| 525 | "offset_bytes": 1017744, |
| 526 | "nbytes": 19008, |
| 527 | "sha256": "31badb447b1ff523651d5544b778714e241515e1db37b04ef08aa0c72d8bba2c" |
| 528 | }, |
| 529 | { |
| 530 | "name": "output.bias", |
| 531 | "shape": [ |
| 532 | 192 |
| 533 | ], |
| 534 | "dtype": "float16", |
| 535 | "offset_bytes": 1036752, |
| 536 | "nbytes": 384, |
| 537 | "sha256": "f14dd4fc8d81b1c497d53b953256be8f2e0a5cf346ffa4cff98ec73e9c504298" |
| 538 | }, |
| 539 | { |
| 540 | "name": "output.weight", |
| 541 | "shape": [ |
| 542 | 192, |
| 543 | 96, |
| 544 | 1 |
| 545 | ], |
| 546 | "dtype": "float16", |
| 547 | "offset_bytes": 1037136, |
| 548 | "nbytes": 36864, |
| 549 | "sha256": "6c771bbca317989d77a2a44a5ae67db14f7a18e0c1cfdd889b9e996085e3a682" |
| 550 | }, |
| 551 | { |
| 552 | "name": "token_blocks.0.net.0.bias", |
| 553 | "shape": [ |
| 554 | 96 |
| 555 | ], |
| 556 | "dtype": "float16", |
| 557 | "offset_bytes": 1074000, |
| 558 | "nbytes": 192, |
| 559 | "sha256": "2a95fc7a53c4a3dd0d9709f117b963365f28e949b83c14ef8d827ee58821874c" |
| 560 | }, |
| 561 | { |
| 562 | "name": "token_blocks.0.net.0.weight", |
| 563 | "shape": [ |
| 564 | 96, |
| 565 | 96, |
| 566 | 5 |
| 567 | ], |
| 568 | "dtype": "float16", |
| 569 | "offset_bytes": 1074192, |
| 570 | "nbytes": 92160, |
| 571 | "sha256": "4466e686498f37dc180d98a905d0aa744a5447a696a2e520347051e1d9e3a853" |
| 572 | }, |
| 573 | { |
| 574 | "name": "token_blocks.0.net.2.bias", |
| 575 | "shape": [ |
| 576 | 96 |
| 577 | ], |
| 578 | "dtype": "float16", |
| 579 | "offset_bytes": 1166352, |
| 580 | "nbytes": 192, |
| 581 | "sha256": "03d74747c7be70419600e9b9a6e4b74b53a522c3ec171dbd9073b967bbed4385" |
| 582 | }, |
| 583 | { |
| 584 | "name": "token_blocks.0.net.2.weight", |
| 585 | "shape": [ |
| 586 | 96, |
| 587 | 96, |
| 588 | 5 |
| 589 | ], |
| 590 | "dtype": "float16", |
| 591 | "offset_bytes": 1166544, |
| 592 | "nbytes": 92160, |
| 593 | "sha256": "06265580330a7252f417ef22db2a29a9f94d117bebea91b7c2da1ce172c99978" |
| 594 | }, |
| 595 | { |
| 596 | "name": "token_blocks.0.scale", |
| 597 | "shape": [ |
| 598 | 1 |
| 599 | ], |
| 600 | "dtype": "float16", |
| 601 | "offset_bytes": 1258704, |
| 602 | "nbytes": 2, |
| 603 | "sha256": "d84e727959ff7628e4773ef6c40b9ae207850fa6863ae4b69bd4eb5757b70c2c" |
| 604 | }, |
| 605 | { |
| 606 | "name": "token_blocks.1.net.0.bias", |
| 607 | "shape": [ |
| 608 | 96 |
| 609 | ], |
| 610 | "dtype": "float16", |
| 611 | "offset_bytes": 1258706, |
| 612 | "nbytes": 192, |
| 613 | "sha256": "e35ce6a21c1474bdbb1f216eac660300d130aefea8e1195b806953cb303a3796" |
| 614 | }, |
| 615 | { |
| 616 | "name": "token_blocks.1.net.0.weight", |
| 617 | "shape": [ |
| 618 | 96, |
| 619 | 96, |
| 620 | 5 |
| 621 | ], |
| 622 | "dtype": "float16", |
| 623 | "offset_bytes": 1258898, |
| 624 | "nbytes": 92160, |
| 625 | "sha256": "b85dbbfe91979a5ee0a7f3401a2e83195e2a1545b04ba3c6689f997d9e203776" |
| 626 | }, |
| 627 | { |
| 628 | "name": "token_blocks.1.net.2.bias", |
| 629 | "shape": [ |
| 630 | 96 |
| 631 | ], |
| 632 | "dtype": "float16", |
| 633 | "offset_bytes": 1351058, |
| 634 | "nbytes": 192, |
| 635 | "sha256": "8d496189006e98b9dba40af4cfa8279b6fb6f2b2fcba686c76e91df13dda0a1a" |
| 636 | }, |
| 637 | { |
| 638 | "name": "token_blocks.1.net.2.weight", |
| 639 | "shape": [ |
| 640 | 96, |
| 641 | 96, |
| 642 | 5 |
| 643 | ], |
| 644 | "dtype": "float16", |
| 645 | "offset_bytes": 1351250, |
| 646 | "nbytes": 92160, |
| 647 | "sha256": "2f673ecc05483267c4f24bf193085ed3bbbd4f5fa527f9600bba6b9bd5eb495a" |
| 648 | }, |
| 649 | { |
| 650 | "name": "token_blocks.1.scale", |
| 651 | "shape": [ |
| 652 | 1 |
| 653 | ], |
| 654 | "dtype": "float16", |
| 655 | "offset_bytes": 1443410, |
| 656 | "nbytes": 2, |
| 657 | "sha256": "0801b810c6df7ea4474d096b362838b41af4ddb1534d364a7c595e9bd3c5e31d" |
| 658 | }, |
| 659 | { |
| 660 | "name": "token_blocks.2.net.0.bias", |
| 661 | "shape": [ |
| 662 | 96 |
| 663 | ], |
| 664 | "dtype": "float16", |
| 665 | "offset_bytes": 1443412, |
| 666 | "nbytes": 192, |
| 667 | "sha256": "40ce499431d265843f4e50d353fa6fce29758596d8cd5da77524995c92640abb" |
| 668 | }, |
| 669 | { |
| 670 | "name": "token_blocks.2.net.0.weight", |
| 671 | "shape": [ |
| 672 | 96, |
| 673 | 96, |
| 674 | 5 |
| 675 | ], |
| 676 | "dtype": "float16", |
| 677 | "offset_bytes": 1443604, |
| 678 | "nbytes": 92160, |
| 679 | "sha256": "0375854a080bc5d1cc79ae1193444772425606dc6d2e7c8897c9e259eb18ef2b" |
| 680 | }, |
| 681 | { |
| 682 | "name": "token_blocks.2.net.2.bias", |
| 683 | "shape": [ |
| 684 | 96 |
| 685 | ], |
| 686 | "dtype": "float16", |
| 687 | "offset_bytes": 1535764, |
| 688 | "nbytes": 192, |
| 689 | "sha256": "a2b404e28598d356f704815906e7115601e5393509732ea51b28afa32656fb83" |
| 690 | }, |
| 691 | { |
| 692 | "name": "token_blocks.2.net.2.weight", |
| 693 | "shape": [ |
| 694 | 96, |
| 695 | 96, |
| 696 | 5 |
| 697 | ], |
| 698 | "dtype": "float16", |
| 699 | "offset_bytes": 1535956, |
| 700 | "nbytes": 92160, |
| 701 | "sha256": "372f57ecc136d3dddbcd59a89a862a7634b1901d6328c0761c19bd699ea2983b" |
| 702 | }, |
| 703 | { |
| 704 | "name": "token_blocks.2.scale", |
| 705 | "shape": [ |
| 706 | 1 |
| 707 | ], |
| 708 | "dtype": "float16", |
| 709 | "offset_bytes": 1628116, |
| 710 | "nbytes": 2, |
| 711 | "sha256": "0848b83f8d5f92005ee25d5bbf610d2f5cbede603455502451c9f96bee736fa3" |
| 712 | }, |
| 713 | { |
| 714 | "name": "token_input_proj.bias", |
| 715 | "shape": [ |
| 716 | 96 |
| 717 | ], |
| 718 | "dtype": "float16", |
| 719 | "offset_bytes": 1628118, |
| 720 | "nbytes": 192, |
| 721 | "sha256": "91801bb10ec9a7224dbb3d19c254885e7a439f0da09af59f22078610bf27e7f5" |
| 722 | }, |
| 723 | { |
| 724 | "name": "token_input_proj.weight", |
| 725 | "shape": [ |
| 726 | 96, |
| 727 | 98, |
| 728 | 1 |
| 729 | ], |
| 730 | "dtype": "float16", |
| 731 | "offset_bytes": 1628310, |
| 732 | "nbytes": 18816, |
| 733 | "sha256": "967cc3300ef7a0930f466ffc4b32b929c2e46a772569d7006fc337fc45e07ce7" |
| 734 | } |
| 735 | ] |
| 736 | }, |
| 737 | "decoder": { |
| 738 | "checkpoint": "artifacts/voices/zh_CN/joint/decoder-student.pt", |
| 739 | "checkpoint_sha256": "82a35ec8d3f445504a1e12d8556666c92dd34338bd8236c1b1d41d478446b749", |
| 740 | "config": { |
| 741 | "in_channels": 192, |
| 742 | "source_in_channels": 192, |
| 743 | "channels": [ |
| 744 | 160, |
| 745 | 80, |
| 746 | 40, |
| 747 | 20 |
| 748 | ], |
| 749 | "res_layers": 1, |
| 750 | "variant": "piperlite", |
| 751 | "lrc_code_dim": 40, |
| 752 | "lrc_encoder_hidden": 64, |
| 753 | "rank_ratio": 0.5, |
| 754 | "activation": "leaky_relu", |
| 755 | "stage_affine": false, |
| 756 | "factorized_pre_rank": 0, |
| 757 | "piper_res_factor_rank_ratio": 0.0, |
| 758 | "res_bank_scale_mode": "kept", |
| 759 | "stage0_branches": [ |
| 760 | 0, |
| 761 | 1, |
| 762 | 2 |
| 763 | ], |
| 764 | "stage1_branches": [ |
| 765 | 0, |
| 766 | 1, |
| 767 | 2 |
| 768 | ], |
| 769 | "stage2_branches": [ |
| 770 | 0, |
| 771 | 1, |
| 772 | 2 |
| 773 | ], |
| 774 | "stage3_branches": [ |
| 775 | 0, |
| 776 | 1, |
| 777 | 2 |
| 778 | ], |
| 779 | "fsd_dim": 72, |
| 780 | "fsd_blocks": 5, |
| 781 | "fsd_film_rank": 12, |
| 782 | "fsd_head_rank": 48, |
| 783 | "wavehax_channels": 32, |
| 784 | "wavehax_blocks": 8, |
| 785 | "wavehax_mult_channels": 2, |
| 786 | "wavehax_kernel_freq": 7, |
| 787 | "wavehax_kernel_time": 7, |
| 788 | "wavehax_f0_channel": 80, |
| 789 | "wavehax_voiced_channel": 81, |
| 790 | "wavehax_sample_rate": 24000, |
| 791 | "wavehax_prior_power": 0.1, |
| 792 | "wavehax_prior_noise": 0.01, |
| 793 | "whx_channels": 32, |
| 794 | "whx_hidden": 64, |
| 795 | "whx_blocks": 8, |
| 796 | "whx_n_fft": 512, |
| 797 | "whx_f0_channel": 80, |
| 798 | "whx_voiced_channel": 81, |
| 799 | "whx_sample_rate": 24000, |
| 800 | "hift_width": 224, |
| 801 | "hift_gen_channels": 192, |
| 802 | "hift_style_dim": 64, |
| 803 | "hift_body_blocks": 3, |
| 804 | "hift_f0_channel": 80, |
| 805 | "hift_voiced_channel": 81, |
| 806 | "hift_sample_rate": 24000, |
| 807 | "stage_projection_bottlenecks": [], |
| 808 | "teacher_init_checkpoint": null, |
| 809 | "teacher_init_method": "first", |
| 810 | "teacher_init": null, |
| 811 | "init_decoder_checkpoint": "artifacts/voices/zh_CN/decoder/decoder-student.pt", |
| 812 | "decoder_init": { |
| 813 | "checkpoint": "artifacts/voices/zh_CN/decoder/decoder-student.pt", |
| 814 | "decoder_parameters": 730721, |
| 815 | "matched_config": { |
| 816 | "in_channels": 192, |
| 817 | "channels": [ |
| 818 | 160, |
| 819 | 80, |
| 820 | 40, |
| 821 | 20 |
| 822 | ], |
| 823 | "res_layers": 1, |
| 824 | "variant": "piperlite", |
| 825 | "rank_ratio": 0.5, |
| 826 | "activation": "leaky_relu", |
| 827 | "stage_affine": false, |
| 828 | "factorized_pre_rank": 0, |
| 829 | "piper_res_factor_rank_ratio": 0.0, |
| 830 | "res_bank_scale_mode": "kept", |
| 831 | "stage0_branches": [ |
| 832 | 0, |
| 833 | 1, |
| 834 | 2 |
| 835 | ], |
| 836 | "stage1_branches": [ |
| 837 | 0, |
| 838 | 1, |
| 839 | 2 |
| 840 | ], |
| 841 | "stage2_branches": [ |
| 842 | 0, |
| 843 | 1, |
| 844 | 2 |
| 845 | ], |
| 846 | "stage3_branches": [ |
| 847 | 0, |
| 848 | 1, |
| 849 | 2 |
| 850 | ], |
| 851 | "fsd_dim": 72, |
| 852 | "fsd_blocks": 5, |
| 853 | "fsd_film_rank": 12, |
| 854 | "fsd_head_rank": 48, |
| 855 | "wavehax_channels": 32, |
| 856 | "wavehax_blocks": 8, |
| 857 | "wavehax_mult_channels": 2, |
| 858 | "wavehax_kernel_freq": 7, |
| 859 | "wavehax_kernel_time": 7, |
| 860 | "wavehax_f0_channel": 80, |
| 861 | "wavehax_voiced_channel": 81, |
| 862 | "wavehax_sample_rate": 24000, |
| 863 | "wavehax_prior_power": 0.1, |
| 864 | "wavehax_prior_noise": 0.01, |
| 865 | "whx_channels": 32, |
| 866 | "whx_hidden": 64, |
| 867 | "whx_blocks": 8, |
| 868 | "whx_n_fft": 512, |
| 869 | "whx_f0_channel": 80, |
| 870 | "whx_voiced_channel": 81, |
| 871 | "whx_sample_rate": 24000, |
| 872 | "hift_width": 224, |
| 873 | "hift_gen_channels": 192, |
| 874 | "hift_style_dim": 64, |
| 875 | "hift_body_blocks": 3, |
| 876 | "hift_f0_channel": 80, |
| 877 | "hift_voiced_channel": 81, |
| 878 | "hift_sample_rate": 24000, |
| 879 | "stage_projection_bottlenecks": [], |
| 880 | "post_filter_channels": 0, |
| 881 | "post_filter_layers": 0, |
| 882 | "post_filter_kernel": 9, |
| 883 | "post_filter_scale": 0.25, |
| 884 | "pre_tanh_repair_channels": 0, |
| 885 | "pre_tanh_repair_layers": 0, |
| 886 | "pre_tanh_repair_kernel": 7, |
| 887 | "pre_tanh_repair_scale": 0.15, |
| 888 | "istft_n_fft": 512, |
| 889 | "istft_tail_head_rank": 20 |
| 890 | }, |
| 891 | "allowed_post_filter_mismatches": [], |
| 892 | "allowed_pre_tanh_repair_mismatches": [], |
| 893 | "allowed_activation_mismatches": [], |
| 894 | "missing_keys": [], |
| 895 | "unexpected_keys": [] |
| 896 | }, |
| 897 | "post_filter_channels": 0, |
| 898 | "post_filter_layers": 0, |
| 899 | "post_filter_kernel": 9, |
| 900 | "post_filter_scale": 0.25, |
| 901 | "pre_tanh_repair_channels": 0, |
| 902 | "pre_tanh_repair_layers": 0, |
| 903 | "pre_tanh_repair_kernel": 7, |
| 904 | "pre_tanh_repair_scale": 0.15, |
| 905 | "istft_n_fft": 512, |
| 906 | "istft_tail_head_rank": 20, |
| 907 | "ap_amplitude_weight": 0.0, |
| 908 | "ap_phase_weight": 0.0, |
| 909 | "ap_complex_weight": 0.0, |
| 910 | "spectral_head_init": "default", |
| 911 | "spectral_head_init_scale": 0.001, |
| 912 | "ap_amp_init_bias": -6.0, |
| 913 | "ap_phase_real_init_bias": 1.0, |
| 914 | "spectral_head_init_summary": null, |
| 915 | "feature_exact_weight": 0.0, |
| 916 | "feature_exact_keys": [ |
| 917 | "pre", |
| 918 | "up0" |
| 919 | ], |
| 920 | "feature_exact_teacher_init": null, |
| 921 | "signature_pack_dir": "artifacts/voices/zh_CN/signatures", |
| 922 | "exact_feature_pack_dir": null, |
| 923 | "signature_hint_weight": 0.05, |
| 924 | "signature_temporal_weight": 0.4, |
| 925 | "signature_phase_weight": 0.0, |
| 926 | "signature_phase_bins": 0, |
| 927 | "signature_keys": [ |
| 928 | "stage0_mix", |
| 929 | "stage1_mix", |
| 930 | "stage2_mix", |
| 931 | "pre_tanh", |
| 932 | "audio" |
| 933 | ], |
| 934 | "bottleneck_code_checkpoint": null, |
| 935 | "bottleneck_code_weight": 0.0, |
| 936 | "bottleneck_code_keys": [], |
| 937 | "bottleneck_code_training_only_parameters": 0, |
| 938 | "echo_tail_weight": 0.0, |
| 939 | "echo_tail_min_ms": 8.0, |
| 940 | "echo_tail_max_ms": 40.0, |
| 941 | "echo_tail_lags": 9, |
| 942 | "echo_tail_margin": 0.02, |
| 943 | "paired_acoustic_residual_weight": 0.0, |
| 944 | "paired_acoustic_residual_max_scale": 0.25, |
| 945 | "lrc_pred_code_mix_prob": 0.0, |
| 946 | "lrc_pred_code_checkpoint": null, |
| 947 | "lrc_pred_code_residual_prob": 0.0, |
| 948 | "lrc_pred_code_residual_max_scale": 0.25, |
| 949 | "lrc_pred_code_cache": null |
| 950 | }, |
| 951 | "parameters": 730721, |
| 952 | "tensors": [ |
| 953 | { |
| 954 | "name": "post.bias", |
| 955 | "shape": [ |
| 956 | 1 |
| 957 | ], |
| 958 | "dtype": "float16", |
| 959 | "offset_bytes": 1647126, |
| 960 | "nbytes": 2, |
| 961 | "sha256": "4491461c9aaa3786193f0e80b1018a44dc5f67bda4637bd26184a2950f1cf852" |
| 962 | }, |
| 963 | { |
| 964 | "name": "post.weight", |
| 965 | "shape": [ |
| 966 | 1, |
| 967 | 20, |
| 968 | 7 |
| 969 | ], |
| 970 | "dtype": "float16", |
| 971 | "offset_bytes": 1647128, |
| 972 | "nbytes": 280, |
| 973 | "sha256": "73b25244ab20974bc954584fabb9f347dd154f72bbabf7550f7f01d04e29d2ea" |
| 974 | }, |
| 975 | { |
| 976 | "name": "pre.bias", |
| 977 | "shape": [ |
| 978 | 160 |
| 979 | ], |
| 980 | "dtype": "float16", |
| 981 | "offset_bytes": 1647408, |
| 982 | "nbytes": 320, |
| 983 | "sha256": "198418d9c92fa852a78e597e6060becade78e77b75f6df6084836f5c4655ed67" |
| 984 | }, |
| 985 | { |
| 986 | "name": "pre.weight", |
| 987 | "shape": [ |
| 988 | 160, |
| 989 | 192, |
| 990 | 7 |
| 991 | ], |
| 992 | "dtype": "float16", |
| 993 | "offset_bytes": 1647728, |
| 994 | "nbytes": 430080, |
| 995 | "sha256": "1670030816f220b6e687f43331722eeb48573229c08873d1a60446bc57137843" |
| 996 | }, |
| 997 | { |
| 998 | "name": "res0.0.blocks.0.conv1.bias", |
| 999 | "shape": [ |
| 1000 | 80 |
| 1001 | ], |
| 1002 | "dtype": "float16", |
| 1003 | "offset_bytes": 2077808, |
| 1004 | "nbytes": 160, |
| 1005 | "sha256": "a85c5b907f650ef6079371b6167007ba309d864319a62623325ee4cc9544d49a" |
| 1006 | }, |
| 1007 | { |
| 1008 | "name": "res0.0.blocks.0.conv1.weight", |
| 1009 | "shape": [ |
| 1010 | 80, |
| 1011 | 80, |
| 1012 | 3 |
| 1013 | ], |
| 1014 | "dtype": "float16", |
| 1015 | "offset_bytes": 2077968, |
| 1016 | "nbytes": 38400, |
| 1017 | "sha256": "4635b57c009e2c18417a30cebc85d3e6bdc7f75e2bb46411ec8245b52882c3cd" |
| 1018 | }, |
| 1019 | { |
| 1020 | "name": "res0.0.blocks.0.conv2.bias", |
| 1021 | "shape": [ |
| 1022 | 80 |
| 1023 | ], |
| 1024 | "dtype": "float16", |
| 1025 | "offset_bytes": 2116368, |
| 1026 | "nbytes": 160, |
| 1027 | "sha256": "4e21f1c5e66e2d70d3b5eac05cccc4d5915692acc774166dad7bf7e02890cfe1" |
| 1028 | }, |
| 1029 | { |
| 1030 | "name": "res0.0.blocks.0.conv2.weight", |
| 1031 | "shape": [ |
| 1032 | 80, |
| 1033 | 80, |
| 1034 | 3 |
| 1035 | ], |
| 1036 | "dtype": "float16", |
| 1037 | "offset_bytes": 2116528, |
| 1038 | "nbytes": 38400, |
| 1039 | "sha256": "0ca3800f6a70ed933acba3ed44b6109c1dd8516903cfd9896082685f357cdf7d" |
| 1040 | }, |
| 1041 | { |
| 1042 | "name": "res0.0.blocks.1.conv1.bias", |
| 1043 | "shape": [ |
| 1044 | 80 |
| 1045 | ], |
| 1046 | "dtype": "float16", |
| 1047 | "offset_bytes": 2154928, |
| 1048 | "nbytes": 160, |
| 1049 | "sha256": "1c6e39777532eb753b4e1583bb8347299aea37c8efbc5b34d34e5fcf0957cb75" |
| 1050 | }, |
| 1051 | { |
| 1052 | "name": "res0.0.blocks.1.conv1.weight", |
| 1053 | "shape": [ |
| 1054 | 80, |
| 1055 | 80, |
| 1056 | 5 |
| 1057 | ], |
| 1058 | "dtype": "float16", |
| 1059 | "offset_bytes": 2155088, |
| 1060 | "nbytes": 64000, |
| 1061 | "sha256": "6d7d576c814e34990a053385cfed1185d1d36c07ec3ac5abefc5c0d27f785f4e" |
| 1062 | }, |
| 1063 | { |
| 1064 | "name": "res0.0.blocks.1.conv2.bias", |
| 1065 | "shape": [ |
| 1066 | 80 |
| 1067 | ], |
| 1068 | "dtype": "float16", |
| 1069 | "offset_bytes": 2219088, |
| 1070 | "nbytes": 160, |
| 1071 | "sha256": "3173c6a7722f5ce86a3663fdbc06544f24f6240698adf323a4f26e5fa4cf261c" |
| 1072 | }, |
| 1073 | { |
| 1074 | "name": "res0.0.blocks.1.conv2.weight", |
| 1075 | "shape": [ |
| 1076 | 80, |
| 1077 | 80, |
| 1078 | 5 |
| 1079 | ], |
| 1080 | "dtype": "float16", |
| 1081 | "offset_bytes": 2219248, |
| 1082 | "nbytes": 64000, |
| 1083 | "sha256": "6a4be39985a6a3cd65918c10cffee414131c881bb796d3bca45c718d82e8920a" |
| 1084 | }, |
| 1085 | { |
| 1086 | "name": "res0.0.blocks.2.conv1.bias", |
| 1087 | "shape": [ |
| 1088 | 80 |
| 1089 | ], |
| 1090 | "dtype": "float16", |
| 1091 | "offset_bytes": 2283248, |
| 1092 | "nbytes": 160, |
| 1093 | "sha256": "c0932534412e26255ade0ee01da2b6923bb32a5aba5a1b386429642f811ef895" |
| 1094 | }, |
| 1095 | { |
| 1096 | "name": "res0.0.blocks.2.conv1.weight", |
| 1097 | "shape": [ |
| 1098 | 80, |
| 1099 | 80, |
| 1100 | 7 |
| 1101 | ], |
| 1102 | "dtype": "float16", |
| 1103 | "offset_bytes": 2283408, |
| 1104 | "nbytes": 89600, |
| 1105 | "sha256": "4fa26cf76e8920ec4ef50b46b4ab1b09967e54aecf3dd11fabae7281a8657127" |
| 1106 | }, |
| 1107 | { |
| 1108 | "name": "res0.0.blocks.2.conv2.bias", |
| 1109 | "shape": [ |
| 1110 | 80 |
| 1111 | ], |
| 1112 | "dtype": "float16", |
| 1113 | "offset_bytes": 2373008, |
| 1114 | "nbytes": 160, |
| 1115 | "sha256": "cb7eede7a725537be3021bcf7c78097a7945b099d66b074a962641d0bdecd402" |
| 1116 | }, |
| 1117 | { |
| 1118 | "name": "res0.0.blocks.2.conv2.weight", |
| 1119 | "shape": [ |
| 1120 | 80, |
| 1121 | 80, |
| 1122 | 7 |
| 1123 | ], |
| 1124 | "dtype": "float16", |
| 1125 | "offset_bytes": 2373168, |
| 1126 | "nbytes": 89600, |
| 1127 | "sha256": "c379706fb1534cefc7cc7f1f2190467db0118864eb0b08976d5ffe31c410988d" |
| 1128 | }, |
| 1129 | { |
| 1130 | "name": "res1.0.blocks.0.conv1.bias", |
| 1131 | "shape": [ |
| 1132 | 40 |
| 1133 | ], |
| 1134 | "dtype": "float16", |
| 1135 | "offset_bytes": 2462768, |
| 1136 | "nbytes": 80, |
| 1137 | "sha256": "fc20be8af0d5184f89b39dc60399e6533e892e13fed1cbaee9c49560c3268556" |
| 1138 | }, |
| 1139 | { |
| 1140 | "name": "res1.0.blocks.0.conv1.weight", |
| 1141 | "shape": [ |
| 1142 | 40, |
| 1143 | 40, |
| 1144 | 3 |
| 1145 | ], |
| 1146 | "dtype": "float16", |
| 1147 | "offset_bytes": 2462848, |
| 1148 | "nbytes": 9600, |
| 1149 | "sha256": "119dfb514d6d3fd3aaf5057145a5b7036c81689cc26f700d4cd0a733ff6a54f8" |
| 1150 | }, |
| 1151 | { |
| 1152 | "name": "res1.0.blocks.0.conv2.bias", |
| 1153 | "shape": [ |
| 1154 | 40 |
| 1155 | ], |
| 1156 | "dtype": "float16", |
| 1157 | "offset_bytes": 2472448, |
| 1158 | "nbytes": 80, |
| 1159 | "sha256": "fbf8e098f116e7f466da16b60bf71023c6b0352ac67c9af27428e1481d511cad" |
| 1160 | }, |
| 1161 | { |
| 1162 | "name": "res1.0.blocks.0.conv2.weight", |
| 1163 | "shape": [ |
| 1164 | 40, |
| 1165 | 40, |
| 1166 | 3 |
| 1167 | ], |
| 1168 | "dtype": "float16", |
| 1169 | "offset_bytes": 2472528, |
| 1170 | "nbytes": 9600, |
| 1171 | "sha256": "c3f0fa64957c8b8384b766eb7e183b8c865312f5571818bc0a139af500c2425a" |
| 1172 | }, |
| 1173 | { |
| 1174 | "name": "res1.0.blocks.1.conv1.bias", |
| 1175 | "shape": [ |
| 1176 | 40 |
| 1177 | ], |
| 1178 | "dtype": "float16", |
| 1179 | "offset_bytes": 2482128, |
| 1180 | "nbytes": 80, |
| 1181 | "sha256": "f175162c2619b532091df8d550c5731bf9a023f50a21571529dc3885e19f3310" |
| 1182 | }, |
| 1183 | { |
| 1184 | "name": "res1.0.blocks.1.conv1.weight", |
| 1185 | "shape": [ |
| 1186 | 40, |
| 1187 | 40, |
| 1188 | 5 |
| 1189 | ], |
| 1190 | "dtype": "float16", |
| 1191 | "offset_bytes": 2482208, |
| 1192 | "nbytes": 16000, |
| 1193 | "sha256": "2775c792017a2d72905c33eeb53ce16e89288daace6c544d9bca92c4b9226812" |
| 1194 | }, |
| 1195 | { |
| 1196 | "name": "res1.0.blocks.1.conv2.bias", |
| 1197 | "shape": [ |
| 1198 | 40 |
| 1199 | ], |
| 1200 | "dtype": "float16", |
| 1201 | "offset_bytes": 2498208, |
| 1202 | "nbytes": 80, |
| 1203 | "sha256": "bf493468d02dedf3ce2309ce4368caf59dc11bcac37b005b8a430db0632c2de6" |
| 1204 | }, |
| 1205 | { |
| 1206 | "name": "res1.0.blocks.1.conv2.weight", |
| 1207 | "shape": [ |
| 1208 | 40, |
| 1209 | 40, |
| 1210 | 5 |
| 1211 | ], |
| 1212 | "dtype": "float16", |
| 1213 | "offset_bytes": 2498288, |
| 1214 | "nbytes": 16000, |
| 1215 | "sha256": "4e14eb8d0e5dfb75e16e26e59580422c1d9d7fe494f255d283676b93b7737a90" |
| 1216 | }, |
| 1217 | { |
| 1218 | "name": "res1.0.blocks.2.conv1.bias", |
| 1219 | "shape": [ |
| 1220 | 40 |
| 1221 | ], |
| 1222 | "dtype": "float16", |
| 1223 | "offset_bytes": 2514288, |
| 1224 | "nbytes": 80, |
| 1225 | "sha256": "12731e2f40a27d23b1227f01badfc3edfa0e091841afad2c6964258feefb7921" |
| 1226 | }, |
| 1227 | { |
| 1228 | "name": "res1.0.blocks.2.conv1.weight", |
| 1229 | "shape": [ |
| 1230 | 40, |
| 1231 | 40, |
| 1232 | 7 |
| 1233 | ], |
| 1234 | "dtype": "float16", |
| 1235 | "offset_bytes": 2514368, |
| 1236 | "nbytes": 22400, |
| 1237 | "sha256": "c586570e5a31d8d8d76d6e90afc83bc802495243a26ddf6f4dbf45c33c2654f2" |
| 1238 | }, |
| 1239 | { |
| 1240 | "name": "res1.0.blocks.2.conv2.bias", |
| 1241 | "shape": [ |
| 1242 | 40 |
| 1243 | ], |
| 1244 | "dtype": "float16", |
| 1245 | "offset_bytes": 2536768, |
| 1246 | "nbytes": 80, |
| 1247 | "sha256": "23897cd7b57f2839723530b524094108d01c892c86e384eb4da978a37689d853" |
| 1248 | }, |
| 1249 | { |
| 1250 | "name": "res1.0.blocks.2.conv2.weight", |
| 1251 | "shape": [ |
| 1252 | 40, |
| 1253 | 40, |
| 1254 | 7 |
| 1255 | ], |
| 1256 | "dtype": "float16", |
| 1257 | "offset_bytes": 2536848, |
| 1258 | "nbytes": 22400, |
| 1259 | "sha256": "af325d5e7379677af9113f7afea69edd168e0d34b03bd595e4c689e49f596bcb" |
| 1260 | }, |
| 1261 | { |
| 1262 | "name": "res2.0.blocks.0.conv1.bias", |
| 1263 | "shape": [ |
| 1264 | 20 |
| 1265 | ], |
| 1266 | "dtype": "float16", |
| 1267 | "offset_bytes": 2559248, |
| 1268 | "nbytes": 40, |
| 1269 | "sha256": "85b58dba90a4abdc1b071401dcb45c98ef4198889e4857f496b71f9da502c0d4" |
| 1270 | }, |
| 1271 | { |
| 1272 | "name": "res2.0.blocks.0.conv1.weight", |
| 1273 | "shape": [ |
| 1274 | 20, |
| 1275 | 20, |
| 1276 | 3 |
| 1277 | ], |
| 1278 | "dtype": "float16", |
| 1279 | "offset_bytes": 2559288, |
| 1280 | "nbytes": 2400, |
| 1281 | "sha256": "9a761f0e93f602c665b0b177ab09d399835ad8b1bb8a2742ce0a91ddd2779916" |
| 1282 | }, |
| 1283 | { |
| 1284 | "name": "res2.0.blocks.0.conv2.bias", |
| 1285 | "shape": [ |
| 1286 | 20 |
| 1287 | ], |
| 1288 | "dtype": "float16", |
| 1289 | "offset_bytes": 2561688, |
| 1290 | "nbytes": 40, |
| 1291 | "sha256": "4f7a80f7edd4976692fcb746b62cd2a60455054e3b2993e258340780ea99a003" |
| 1292 | }, |
| 1293 | { |
| 1294 | "name": "res2.0.blocks.0.conv2.weight", |
| 1295 | "shape": [ |
| 1296 | 20, |
| 1297 | 20, |
| 1298 | 3 |
| 1299 | ], |
| 1300 | "dtype": "float16", |
| 1301 | "offset_bytes": 2561728, |
| 1302 | "nbytes": 2400, |
| 1303 | "sha256": "70d602100f6cca7a8885a160323fb75bc9263f102a79475faf7b5643593ace0a" |
| 1304 | }, |
| 1305 | { |
| 1306 | "name": "res2.0.blocks.1.conv1.bias", |
| 1307 | "shape": [ |
| 1308 | 20 |
| 1309 | ], |
| 1310 | "dtype": "float16", |
| 1311 | "offset_bytes": 2564128, |
| 1312 | "nbytes": 40, |
| 1313 | "sha256": "24d73f669e986a692f53615cada4bbf4ef643f32e359747bb185b736800727f5" |
| 1314 | }, |
| 1315 | { |
| 1316 | "name": "res2.0.blocks.1.conv1.weight", |
| 1317 | "shape": [ |
| 1318 | 20, |
| 1319 | 20, |
| 1320 | 5 |
| 1321 | ], |
| 1322 | "dtype": "float16", |
| 1323 | "offset_bytes": 2564168, |
| 1324 | "nbytes": 4000, |
| 1325 | "sha256": "fe1213f84baab8a81c06436db0257c6dbe1da6499378bf492b550dc7e6e5c61e" |
| 1326 | }, |
| 1327 | { |
| 1328 | "name": "res2.0.blocks.1.conv2.bias", |
| 1329 | "shape": [ |
| 1330 | 20 |
| 1331 | ], |
| 1332 | "dtype": "float16", |
| 1333 | "offset_bytes": 2568168, |
| 1334 | "nbytes": 40, |
| 1335 | "sha256": "11fb416c66b3683ae2243b501655ca67c7b7d063213d2d7f5c7597c0f5752071" |
| 1336 | }, |
| 1337 | { |
| 1338 | "name": "res2.0.blocks.1.conv2.weight", |
| 1339 | "shape": [ |
| 1340 | 20, |
| 1341 | 20, |
| 1342 | 5 |
| 1343 | ], |
| 1344 | "dtype": "float16", |
| 1345 | "offset_bytes": 2568208, |
| 1346 | "nbytes": 4000, |
| 1347 | "sha256": "b38c11b3e1c303fa3aa02605292f080957ed2ac41af8bd5b3a0a0e608f65eb12" |
| 1348 | }, |
| 1349 | { |
| 1350 | "name": "res2.0.blocks.2.conv1.bias", |
| 1351 | "shape": [ |
| 1352 | 20 |
| 1353 | ], |
| 1354 | "dtype": "float16", |
| 1355 | "offset_bytes": 2572208, |
| 1356 | "nbytes": 40, |
| 1357 | "sha256": "e3ed1d4683462e8aca530768b5934ed09cb9ff998331e35e8603703ed4af3395" |
| 1358 | }, |
| 1359 | { |
| 1360 | "name": "res2.0.blocks.2.conv1.weight", |
| 1361 | "shape": [ |
| 1362 | 20, |
| 1363 | 20, |
| 1364 | 7 |
| 1365 | ], |
| 1366 | "dtype": "float16", |
| 1367 | "offset_bytes": 2572248, |
| 1368 | "nbytes": 5600, |
| 1369 | "sha256": "41fc0a9a5adf987a50fb13f1106f914e035d75fb7e433bb0e61e4ec7226b0fb1" |
| 1370 | }, |
| 1371 | { |
| 1372 | "name": "res2.0.blocks.2.conv2.bias", |
| 1373 | "shape": [ |
| 1374 | 20 |
| 1375 | ], |
| 1376 | "dtype": "float16", |
| 1377 | "offset_bytes": 2577848, |
| 1378 | "nbytes": 40, |
| 1379 | "sha256": "8982863dbbd1bc452eaa12a7361624e468a7ac6663b25875e858517c34cfb172" |
| 1380 | }, |
| 1381 | { |
| 1382 | "name": "res2.0.blocks.2.conv2.weight", |
| 1383 | "shape": [ |
| 1384 | 20, |
| 1385 | 20, |
| 1386 | 7 |
| 1387 | ], |
| 1388 | "dtype": "float16", |
| 1389 | "offset_bytes": 2577888, |
| 1390 | "nbytes": 5600, |
| 1391 | "sha256": "a519bd609748d51eb99c199e3143ed432fde88938f903122eb5618ed9a6bcb7f" |
| 1392 | }, |
| 1393 | { |
| 1394 | "name": "up0.bias", |
| 1395 | "shape": [ |
| 1396 | 80 |
| 1397 | ], |
| 1398 | "dtype": "float16", |
| 1399 | "offset_bytes": 2583488, |
| 1400 | "nbytes": 160, |
| 1401 | "sha256": "97151360f93c04e570c0d0b7b1635e3a6d3d84914509714289dceda3961f0b0c" |
| 1402 | }, |
| 1403 | { |
| 1404 | "name": "up0.weight", |
| 1405 | "shape": [ |
| 1406 | 160, |
| 1407 | 80, |
| 1408 | 16 |
| 1409 | ], |
| 1410 | "dtype": "float16", |
| 1411 | "offset_bytes": 2583648, |
| 1412 | "nbytes": 409600, |
| 1413 | "sha256": "4d46eb467a7d6026a77de9a4368f09265fc0f6aa63c278ed7ed1b4be7cf62b5b" |
| 1414 | }, |
| 1415 | { |
| 1416 | "name": "up1.bias", |
| 1417 | "shape": [ |
| 1418 | 40 |
| 1419 | ], |
| 1420 | "dtype": "float16", |
| 1421 | "offset_bytes": 2993248, |
| 1422 | "nbytes": 80, |
| 1423 | "sha256": "6d4c3c691b3fd3a2cfd9e3dfd2baf2ac020d4d82f38802475edea3173f21f492" |
| 1424 | }, |
| 1425 | { |
| 1426 | "name": "up1.weight", |
| 1427 | "shape": [ |
| 1428 | 80, |
| 1429 | 40, |
| 1430 | 16 |
| 1431 | ], |
| 1432 | "dtype": "float16", |
| 1433 | "offset_bytes": 2993328, |
| 1434 | "nbytes": 102400, |
| 1435 | "sha256": "2be36bb4983bd3be7c2a85ee6059b6bfbfa61e2c9f14ac269db7e71c4a46c3c5" |
| 1436 | }, |
| 1437 | { |
| 1438 | "name": "up2.bias", |
| 1439 | "shape": [ |
| 1440 | 20 |
| 1441 | ], |
| 1442 | "dtype": "float16", |
| 1443 | "offset_bytes": 3095728, |
| 1444 | "nbytes": 40, |
| 1445 | "sha256": "580007f1f9049a3d6f358f01169759ae1c8375ae96f79a07621b3fc13bda18d3" |
| 1446 | }, |
| 1447 | { |
| 1448 | "name": "up2.weight", |
| 1449 | "shape": [ |
| 1450 | 40, |
| 1451 | 20, |
| 1452 | 8 |
| 1453 | ], |
| 1454 | "dtype": "float16", |
| 1455 | "offset_bytes": 3095768, |
| 1456 | "nbytes": 12800, |
| 1457 | "sha256": "b4eccdfcfc0b38e3ddba8d470684503ff69985ba808db4e9cf534c18c37ad738" |
| 1458 | } |
| 1459 | ] |
| 1460 | } |
| 1461 | }, |
| 1462 | "runtime_ops": [ |
| 1463 | "embedding_lookup", |
| 1464 | "conv1d_same", |
| 1465 | "conv1d_1x1", |
| 1466 | "conv_transpose1d", |
| 1467 | "silu", |
| 1468 | "leaky_relu", |
| 1469 | "tanh", |
| 1470 | "residual_add", |
| 1471 | "repeat_interleave", |
| 1472 | "masked_conv1d", |
| 1473 | "depthwise_conv1d", |
| 1474 | "low_rank_linear_or_conv1d" |
| 1475 | ], |
| 1476 | "runtime_boundary": { |
| 1477 | "neural_stack_self_contained": true, |
| 1478 | "arbitrary_text_self_contained": false, |
| 1479 | "accepted_runtime_input": "Piper phoneme ID sequence matching piper-phoneme-config.json", |
| 1480 | "output": "mono float waveform at manifest.sample_rate" |
| 1481 | }, |
| 1482 | "frontend": { |
| 1483 | "status": "not fully self-contained for arbitrary text yet", |
| 1484 | "input_contract": "runtime may accept Piper phoneme ID sequences directly", |
| 1485 | "included_config": "piper-phoneme-config.json", |
| 1486 | "piper_phoneme_type": "pinyin", |
| 1487 | "piper_espeak_voice": "zh", |
| 1488 | "phoneme_id_map_entries": 85, |
| 1489 | "next_work": "embed or rewrite the Piper/eSpeak-compatible frontend, or keep a strict phoneme-ID API" |
| 1490 | }, |
| 1491 | "size_budget": { |
| 1492 | "target_max_bytes": 4000000, |
| 1493 | "target_max_mib": 3.814697265625, |
| 1494 | "actual_model_asset_bytes": 3111495, |
| 1495 | "target_max_parameters": 1500000, |
| 1496 | "under_parameter_budget": false, |
| 1497 | "under_4mb_asset_budget": true |
| 1498 | }, |
| 1499 | "limitations": [ |
| 1500 | "This is a raw neural asset package, not a compiled standalone runtime binary.", |
| 1501 | "No PyTorch checkpoint, ONNX graph, or training state is included.", |
| 1502 | "Arbitrary text still requires a Piper/eSpeak-compatible frontend outside this package." |
| 1503 | ] |
| 1504 | } |
| 1505 | |