File size: 9,482 Bytes
ea07244 6781708 ea07244 6781708 994c238 6781708 f0a085b 06b4245 f0a085b 06b4245 f7cc7c3 b26be81 06b4245 f7cc7c3 06b4245 ea07244 b26be81 ea07244 8ebeeb8 bf79b4c ea07244 |
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 |
# Copyright 2022-2023 Xiaomi Corp. (authors: Fangjun Kuang)
#
# See LICENSE for clarification regarding multiple authors
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from functools import lru_cache
import sherpa_onnx
from huggingface_hub import hf_hub_download
sample_rate = 16000
def _get_nn_model_filename(
repo_id: str,
filename: str,
subfolder: str = "exp",
) -> str:
nn_model_filename = hf_hub_download(
repo_id=repo_id,
filename=filename,
subfolder=subfolder,
)
return nn_model_filename
get_file = _get_nn_model_filename
def _get_bpe_model_filename(
repo_id: str,
filename: str = "bpe.model",
subfolder: str = "data/lang_bpe_500",
) -> str:
bpe_model_filename = hf_hub_download(
repo_id=repo_id,
filename=filename,
subfolder=subfolder,
)
return bpe_model_filename
def _get_token_filename(
repo_id: str,
filename: str = "tokens.txt",
subfolder: str = "data/lang_char",
) -> str:
token_filename = hf_hub_download(
repo_id=repo_id,
filename=filename,
subfolder=subfolder,
)
return token_filename
@lru_cache(maxsize=10)
def _get_whisper_model(repo_id: str) -> sherpa_onnx.OfflineRecognizer:
name = repo_id.split("-")[1]
assert name in ("tiny.en", "base.en", "small.en", "medium.en"), repo_id
full_repo_id = "csukuangfj/sherpa-onnx-whisper-" + name
encoder = _get_nn_model_filename(
repo_id=full_repo_id,
filename=f"{name}-encoder.int8.ort",
subfolder=".",
)
decoder = _get_nn_model_filename(
repo_id=full_repo_id,
filename=f"{name}-decoder.int8.ort",
subfolder=".",
)
tokens = _get_token_filename(
repo_id=full_repo_id, subfolder=".", filename=f"{name}-tokens.txt"
)
recognizer = sherpa_onnx.OfflineRecognizer.from_whisper(
encoder=encoder,
decoder=decoder,
tokens=tokens,
num_threads=2,
)
return recognizer
@lru_cache(maxsize=10)
def _get_paraformer_zh_pre_trained_model(repo_id: str) -> sherpa_onnx.OfflineRecognizer:
assert repo_id in [
"csukuangfj/sherpa-onnx-paraformer-zh-2023-03-28",
], repo_id
nn_model = _get_nn_model_filename(
repo_id=repo_id,
filename="model.int8.onnx",
subfolder=".",
)
tokens = _get_token_filename(repo_id=repo_id, subfolder=".")
recognizer = sherpa_onnx.OfflineRecognizer.from_paraformer(
paraformer=nn_model,
tokens=tokens,
num_threads=2,
sample_rate=sample_rate,
feature_dim=80,
decoding_method="greedy_search",
debug=False,
)
return recognizer
@lru_cache(maxsize=10)
def _get_russian_pre_trained_model(repo_id: str) -> sherpa_onnx.OfflineRecognizer:
assert repo_id in (
"alphacep/vosk-model-ru",
"alphacep/vosk-model-small-ru",
), repo_id
if repo_id == "alphacep/vosk-model-ru":
model_dir = "am-onnx"
elif repo_id == "alphacep/vosk-model-small-ru":
model_dir = "am"
encoder_model = _get_nn_model_filename(
repo_id=repo_id,
filename="encoder.onnx",
subfolder=model_dir,
)
decoder_model = _get_nn_model_filename(
repo_id=repo_id,
filename="decoder.onnx",
subfolder=model_dir,
)
joiner_model = _get_nn_model_filename(
repo_id=repo_id,
filename="joiner.onnx",
subfolder=model_dir,
)
tokens = _get_token_filename(repo_id=repo_id, subfolder="lang")
recognizer = sherpa_onnx.OfflineRecognizer.from_transducer(
tokens=tokens,
encoder=encoder_model,
decoder=decoder_model,
joiner=joiner_model,
num_threads=2,
sample_rate=16000,
feature_dim=80,
decoding_method="greedy_search",
)
return recognizer
@lru_cache(maxsize=2)
def get_vad() -> sherpa_onnx.VoiceActivityDetector:
vad_model = _get_nn_model_filename(
repo_id="csukuangfj/vad",
filename="silero_vad.onnx",
subfolder=".",
)
config = sherpa_onnx.VadModelConfig()
config.silero_vad.model = vad_model
config.silero_vad.min_silence_duration = 0.15
config.silero_vad.min_speech_duration = 0.25
config.sample_rate = sample_rate
vad = sherpa_onnx.VoiceActivityDetector(
config,
buffer_size_in_seconds=180,
)
return vad
@lru_cache(maxsize=10)
def get_pretrained_model(repo_id: str) -> sherpa_onnx.OfflineRecognizer:
if repo_id in chinese_models:
return chinese_models[repo_id](repo_id)
elif repo_id in english_models:
return english_models[repo_id](repo_id)
elif repo_id in chinese_english_mixed_models:
return chinese_english_mixed_models[repo_id](repo_id)
elif repo_id in russian_models:
return russian_models[repo_id](repo_id)
else:
raise ValueError(f"Unsupported repo_id: {repo_id}")
def _get_wenetspeech_pre_trained_model(repo_id):
assert repo_id in (
"csukuangfj/sherpa-onnx-conformer-zh-stateless2-2023-05-23",
), repo_id
encoder_model = _get_nn_model_filename(
repo_id=repo_id,
filename="encoder-epoch-99-avg-1.onnx",
subfolder=".",
)
decoder_model = _get_nn_model_filename(
repo_id=repo_id,
filename="decoder-epoch-99-avg-1.onnx",
subfolder=".",
)
joiner_model = _get_nn_model_filename(
repo_id=repo_id,
filename="joiner-epoch-99-avg-1.onnx",
subfolder=".",
)
tokens = _get_token_filename(repo_id=repo_id, subfolder=".")
recognizer = sherpa_onnx.OfflineRecognizer.from_transducer(
tokens=tokens,
encoder=encoder_model,
decoder=decoder_model,
joiner=joiner_model,
num_threads=2,
sample_rate=16000,
feature_dim=80,
decoding_method="greedy_search",
)
return recognizer
def _get_multi_zh_hans_pre_trained_model(repo_id):
assert repo_id in ("zrjin/sherpa-onnx-zipformer-multi-zh-hans-2023-9-2",), repo_id
encoder_model = _get_nn_model_filename(
repo_id=repo_id,
filename="encoder-epoch-20-avg-1.onnx",
subfolder=".",
)
decoder_model = _get_nn_model_filename(
repo_id=repo_id,
filename="decoder-epoch-20-avg-1.onnx",
subfolder=".",
)
joiner_model = _get_nn_model_filename(
repo_id=repo_id,
filename="joiner-epoch-20-avg-1.onnx",
subfolder=".",
)
tokens = _get_token_filename(repo_id=repo_id, subfolder=".")
recognizer = sherpa_onnx.OfflineRecognizer.from_transducer(
tokens=tokens,
encoder=encoder_model,
decoder=decoder_model,
joiner=joiner_model,
num_threads=2,
sample_rate=16000,
feature_dim=80,
decoding_method="greedy_search",
)
return recognizer
def _get_english_model(repo_id: str) -> sherpa_onnx.OfflineRecognizer:
assert (
repo_id
== "yfyeung/icefall-asr-multidataset-pruned_transducer_stateless7-2023-05-04"
), repo_id
encoder_model = _get_nn_model_filename(
repo_id=repo_id,
filename="encoder-epoch-30-avg-4.onnx",
subfolder="exp",
)
decoder_model = _get_nn_model_filename(
repo_id=repo_id,
filename="decoder-epoch-30-avg-4.onnx",
subfolder="exp",
)
joiner_model = _get_nn_model_filename(
repo_id=repo_id,
filename="joiner-epoch-30-avg-4.onnx",
subfolder="exp",
)
tokens = _get_token_filename(repo_id=repo_id, subfolder="lang_bpe_500")
recognizer = sherpa_onnx.OfflineRecognizer.from_transducer(
tokens=tokens,
encoder=encoder_model,
decoder=decoder_model,
joiner=joiner_model,
num_threads=2,
sample_rate=16000,
feature_dim=80,
decoding_method="greedy_search",
)
return recognizer
chinese_models = {
"csukuangfj/sherpa-onnx-conformer-zh-stateless2-2023-05-23": _get_wenetspeech_pre_trained_model, # noqa
"zrjin/sherpa-onnx-zipformer-multi-zh-hans-2023-9-2": _get_multi_zh_hans_pre_trained_model, # noqa
}
english_models = {
"whisper-tiny.en": _get_whisper_model,
"whisper-base.en": _get_whisper_model,
"whisper-small.en": _get_whisper_model,
"yfyeung/icefall-asr-multidataset-pruned_transducer_stateless7-2023-05-04": _get_english_model, # noqa
}
chinese_english_mixed_models = {
"csukuangfj/sherpa-onnx-paraformer-zh-2023-03-28": _get_paraformer_zh_pre_trained_model,
}
russian_models = {
"alphacep/vosk-model-ru": _get_russian_pre_trained_model,
"alphacep/vosk-model-small-ru": _get_russian_pre_trained_model,
}
language_to_models = {
"Chinese+English": list(chinese_english_mixed_models.keys()),
"Chinese": list(chinese_models.keys()),
"English": list(english_models.keys()),
"Russian": list(russian_models.keys()),
}
|