def __init__(
self,
fs: Union[int, str] = 16000,
frontend_conf: Optional[dict] = get_default_kwargs(Frontend),
download_dir: str = None,
multilayer_feature: bool = False,
):
assert check_argument_types()
super().__init__()
if isinstance(fs, str):
fs = humanfriendly.parse_size(fs)
if download_dir is not None:
torch.hub.set_dir(download_dir)
self.multilayer_feature = multilayer_feature
self.upstream, self.featurizer = self._get_upstream(frontend_conf)
if getattr(
self.upstream, "model", None
) is not None and self.upstream.model.__class__.__name__ in [
"Wav2Vec2Model",
"HuberModel",
]:
self.upstream.model.encoder.layerdrop = 0.0
self.pretrained_params = copy.deepcopy(self.upstream.state_dict())
self.output_dim = self.featurizer.output_dim
def enh_s2t_config_file(tmp_path: Path, token_list):
# Write default configuration file
EnhS2TTask.main(
cmd=[
"--dry_run",
"true",
"--output_dir",
str(tmp_path / "enh_s2t"),
"--token_list",
str(token_list),
"--token_type",
"char",
]
)
with open(tmp_path / "enh_s2t" / "config.yaml", "r") as f:
args = yaml.safe_load(f)
if args["enh_encoder"] == "stft" and len(args["enh_encoder_conf"]) == 0:
args["enh_encoder_conf"] = get_default_kwargs(STFTEncoder)
with open(tmp_path / "enh_s2t" / "config.yaml", "w") as f:
yaml_no_alias_safe_dump(args, f, indent=4, sort_keys=False)
return tmp_path / "enh_s2t" / "config.yaml"
def add_task_arguments(cls, parser: argparse.ArgumentParser):
# NOTE(kamo): Use '_' instead of '-' to avoid confusion
assert check_argument_types()
group = parser.add_argument_group(description="Task related")
# NOTE(kamo): add_arguments(..., required=True) can't be used
# to provide --print_config mode. Instead of it, do as
required = parser.get_default("required")
required += ["token_list"]
group.add_argument(
"--token_list",
type=str_or_none,
default=None,
help="A text mapping int-id to token",
)
group.add_argument(
"--odim",
type=int_or_none,
default=None,
help="The number of dimension of output feature",
)
group.add_argument(
"--model_conf",
action=NestedDictAction,
default=get_default_kwargs(ESPnetTTSModel),
help="The keyword arguments for model class.",
)
group = parser.add_argument_group(description="Preprocess related")
group.add_argument(
"--use_preprocessor",
type=str2bool,
default=False,
help="Apply preprocessing to data or not",
)
group.add_argument(
"--token_type",
type=str,
default="bpe",
choices=["bpe", "char", "word"],
help="The text will be tokenized "
"in the specified level token",
)
group.add_argument(
"--bpemodel",
type=str_or_none,
default=None,
help="The model file of sentencepiece",
)
parser.add_argument(
"--non_linguistic_symbols",
type=str_or_none,
help="non_linguistic_symbols file path",
)
for class_choices in cls.class_choices_list:
# Append --<name> and --<name>_conf.
# e.g. --encoder and --encoder_conf
class_choices.add_arguments(group)
def __init__(
self,
fs: Union[int, str] = 16000,
n_fft: int = 512,
win_length: int = None,
hop_length: int = 128,
window: Optional[str] = "hann",
center: bool = True,
normalized: bool = False,
onesided: bool = True,
n_mels: int = 80,
fmin: int = None,
fmax: int = None,
htk: bool = False,
frontend_conf: Optional[dict] = get_default_kwargs(Frontend),
apply_stft: bool = True,
):
assert check_argument_types()
super().__init__()
if isinstance(fs, str):
fs = humanfriendly.parse_size(fs)
# Deepcopy (In general, dict shouldn't be used as default arg)
frontend_conf = copy.deepcopy(frontend_conf)
self.hop_length = hop_length
if apply_stft:
self.stft = Stft(
n_fft=n_fft,
win_length=win_length,
hop_length=hop_length,
center=center,
window=window,
normalized=normalized,
onesided=onesided,
)
else:
self.stft = None
self.apply_stft = apply_stft
if frontend_conf is not None:
self.frontend = Frontend(idim=n_fft // 2 + 1, **frontend_conf)
else:
self.frontend = None
self.logmel = LogMel(
fs=fs,
n_fft=n_fft,
n_mels=n_mels,
fmin=fmin,
fmax=fmax,
htk=htk,
)
self.n_mels = n_mels
self.frontend_type = "default"
def add_task_arguments(cls, parser: argparse.ArgumentParser):
group = parser.add_argument_group(description="Task related")
# NOTE(kamo): add_arguments(..., required=True) can't be used
# to provide --print_config mode. Instead of it, do as
# required = parser.get_default("required")
group.add_argument(
"--init",
type=lambda x: str_or_none(x.lower()),
default=None,
help="The initialization method",
choices=[
"chainer",
"xavier_uniform",
"xavier_normal",
"kaiming_uniform",
"kaiming_normal",
None,
],
)
group.add_argument(
"--model_conf",
action=NestedDictAction,
default=get_default_kwargs(ESPnetEnhancementModel),
help="The keyword arguments for model class.",
)
group.add_argument(
"--criterions",
action=NestedDictAction,
default=[
{
"name": "si_snr",
"conf": {},
"wrapper": "fixed_order",
"wrapper_conf": {},
},
],
help="The criterions binded with the loss wrappers.",
)
group = parser.add_argument_group(description="Preprocess related")
group.add_argument(
"--use_preprocessor",
type=str2bool,
default=False,
help="Apply preprocessing to data or not",
)
for class_choices in cls.class_choices_list:
# Append --<name> and --<name>_conf.
# e.g. --encoder and --encoder_conf
class_choices.add_arguments(group)
def add_task_arguments(cls, parser: argparse.ArgumentParser):
group = parser.add_argument_group(description="Task related")
group.add_argument(
"--num_spk",
type=int_or_none,
default=None,
help=
"The number fo speakers (for each recording) used in system training",
)
group.add_argument(
"--init",
type=lambda x: str_or_none(x.lower()),
default=None,
help="The initialization method",
choices=[
"chainer",
"xavier_uniform",
"xavier_normal",
"kaiming_uniform",
"kaiming_normal",
None,
],
)
group.add_argument(
"--input_size",
type=int_or_none,
default=None,
help="The number of input dimension of the feature",
)
group.add_argument(
"--model_conf",
action=NestedDictAction,
default=get_default_kwargs(ESPnetDiarizationModel),
help="The keyword arguments for model class.",
)
group = parser.add_argument_group(description="Preprocess related")
group.add_argument(
"--use_preprocessor",
type=str2bool,
default=True,
help="Apply preprocessing to data or not",
)
for class_choices in cls.class_choices_list:
# Append --<name> and --<name>_conf.
# e.g. --encoder and --encoder_conf
class_choices.add_arguments(group)
def __init__(
self,
fs: Union[int, str] = 16000,
frontend_conf: Optional[dict] = get_default_kwargs(Frontend),
download_dir: str = None,
multilayer_feature: bool = False,
):
try:
import s3prl
from s3prl.nn import Featurizer, S3PRLUpstream
except Exception as e:
print("Error: S3PRL is not properly installed.")
print(
"Please install S3PRL: cd ${MAIN_ROOT}/tools && make s3prl.done"
)
raise e
assert check_argument_types()
super().__init__()
if isinstance(fs, str):
fs = humanfriendly.parse_size(fs)
if fs != 16000:
logging.warning(
"All the upstream models in S3PRL now only support 16 kHz audio."
)
if download_dir is not None:
s3prl.util.download.set_dir(download_dir)
assert frontend_conf.get("upstream",
None) in S3PRLUpstream.available_names()
upstream = S3PRLUpstream(
frontend_conf.get("upstream"),
path_or_url=frontend_conf.get("path_or_url", None),
)
upstream.eval()
if getattr(upstream, "model",
None) is not None and upstream.model.__class__.__name__ in [
"Wav2Vec2Model",
"HubertModel",
]:
upstream.model.encoder.layerdrop = 0.0
featurizer = Featurizer(upstream)
self.multilayer_feature = multilayer_feature
self.upstream, self.featurizer = upstream, featurizer
self.pretrained_params = copy.deepcopy(self.upstream.state_dict())
self.frontend_type = "s3prl"
self.hop_length = self.featurizer.downsample_rate
self.tile_factor = frontend_conf.get("tile_factor", 1)
def __init__(
self,
fs: Union[int, str] = 16000,
frontend_conf: Optional[dict] = get_default_kwargs(Frontend),
download_dir: str = None,
multilayer_feature: bool = False,
):
assert check_argument_types()
super().__init__()
if isinstance(fs, str):
fs = humanfriendly.parse_size(fs)
if download_dir is not None:
torch.hub.set_dir(download_dir)
self.multilayer_feature = multilayer_feature
self.upstream, self.featurizer = self._get_upstream(frontend_conf)
self.pretrained_params = copy.deepcopy(self.upstream.state_dict())
self.output_dim = self.featurizer.output_dim
def add_task_arguments(cls, parser: argparse.ArgumentParser):
group = parser.add_argument_group(description="Task related")
# NOTE(kamo): add_arguments(..., required=True) can't be used
# to provide --print_config mode. Instead of it, do as
required = parser.get_default("required")
group.add_argument(
"--init",
type=lambda x: str_or_none(x.lower()),
default=None,
help="The initialization method",
choices=[
"chainer",
"xavier_uniform",
"xavier_normal",
"kaiming_uniform",
"kaiming_normal",
None,
],
)
group.add_argument(
"--input_size",
type=int_or_none,
default=None,
help="The number of input dimension of the feature",
)
group.add_argument(
"--model_conf",
action=NestedDictAction,
default=get_default_kwargs(HynetImgrModel),
help="The keyword arguments for model class.",
)
for class_choices in cls.class_choices_list:
# Append --<name> and --<name>_conf.
# e.g. --encoder and --encoder_conf
class_choices.add_arguments(group)
def add_task_arguments(cls, parser: argparse.ArgumentParser):
group = parser.add_argument_group(description="Task related")
# NOTE(kamo): add_arguments(..., required=True) can't be used
# to provide --print_config mode. Instead of it, do as
required = parser.get_default("required")
required += ["token_list"]
group.add_argument(
"--token_list",
type=str_or_none,
default=None,
help="A text mapping int-id to token",
)
group.add_argument(
"--init",
type=lambda x: str_or_none(x.lower()),
default=None,
help="The initialization method",
choices=[
"chainer",
"xavier_uniform",
"xavier_normal",
"kaiming_uniform",
"kaiming_normal",
None,
],
)
group.add_argument(
"--input_size",
type=int_or_none,
default=None,
help="The number of input dimension of the feature",
)
group.add_argument(
"--ctc_conf",
action=NestedDictAction,
default=get_default_kwargs(CTC),
help="The keyword arguments for CTC class.",
)
group.add_argument(
"--model_conf",
action=NestedDictAction,
default=get_default_kwargs(ESPnetASRModel),
help="The keyword arguments for model class.",
)
group.add_argument(
"--use_curriculum",
type=str,
default=False,
help="Use curriculum learning algorithms.",
)
group.add_argument(
"--curriculum_algo",
type=str,
default="exp3s",
help="MAB algorithm to use for curriculum learning",
)
group.add_argument(
"--gain_type",
type=str,
default="PG",
help="Loss-based gain for Curriculum Learning. Prediction Gain (PG) or Self-Prediction Gain (SPG)",
)
group.add_argument("--refill_task",
type=bool,
default=1,
help="Refill task after the task is exhausted")
group.add_argument("--gen_log_dir",
type=str,
default='curriculum_log',
help="Directory to store curriculum generator logs")
group = parser.add_argument_group(description="Preprocess related")
group.add_argument(
"--use_preprocessor",
type=str2bool,
default=True,
help="Apply preprocessing to data or not",
)
group.add_argument(
"--token_type",
type=str,
default="bpe",
choices=["bpe", "char", "word", "phn"],
help="The text will be tokenized " "in the specified level token",
)
group.add_argument(
"--bpemodel",
type=str_or_none,
default=None,
help="The model file of sentencepiece",
)
parser.add_argument(
"--non_linguistic_symbols",
type=str_or_none,
help="non_linguistic_symbols file path",
)
parser.add_argument(
"--cleaner",
type=str_or_none,
choices=[None, "tacotron", "jaconv", "vietnamese"],
default=None,
help="Apply text cleaning",
)
parser.add_argument(
"--g2p",
type=str_or_none,
choices=[None, "g2p_en", "pyopenjtalk", "pyopenjtalk_kana"],
default=None,
help="Specify g2p method if --token_type=phn",
)
parser.add_argument(
"--speech_volume_normalize",
type=float_or_none,
default=None,
help="Scale the maximum amplitude to the given value.",
)
parser.add_argument(
"--rir_scp",
type=str_or_none,
default=None,
help="The file path of rir scp file.",
)
parser.add_argument(
"--rir_apply_prob",
type=float,
default=1.0,
help="THe probability for applying RIR convolution.",
)
parser.add_argument(
"--noise_scp",
type=str_or_none,
default=None,
help="The file path of noise scp file.",
)
parser.add_argument(
"--noise_apply_prob",
type=float,
default=1.0,
help="The probability applying Noise adding.",
)
parser.add_argument(
"--noise_db_range",
type=str,
default="13_15",
help="The range of noise decibel level.",
)
for class_choices in cls.class_choices_list:
# Append --<name> and --<name>_conf.
# e.g. --encoder and --encoder_conf
class_choices.add_arguments(group)
def add_task_arguments(cls, parser: argparse.ArgumentParser):
group = parser.add_argument_group(description="Task related")
group.add_argument(
"--token_list",
type=str_or_none,
default=None,
help="A text mapping int-id to token",
)
group.add_argument(
"--src_token_list",
type=str_or_none,
default=None,
help="A text mapping int-id to token (for source language)",
)
group.add_argument(
"--init",
type=lambda x: str_or_none(x.lower()),
default=None,
help="The initialization method",
choices=[
"chainer",
"xavier_uniform",
"xavier_normal",
"kaiming_uniform",
"kaiming_normal",
None,
],
)
group.add_argument(
"--input_size",
type=int_or_none,
default=None,
help="The number of input dimension of the feature",
)
group.add_argument(
"--ctc_conf",
action=NestedDictAction,
default=get_default_kwargs(CTC),
help="The keyword arguments for CTC class.",
)
group.add_argument(
"--enh_criterions",
action=NestedDictAction,
default=[
{
"name": "si_snr",
"conf": {},
"wrapper": "fixed_order",
"wrapper_conf": {},
},
],
help="The criterions binded with the loss wrappers.",
)
group.add_argument(
"--diar_num_spk",
type=int_or_none,
default=None,
help="The number of speakers (for each recording) for diar submodel class",
)
group.add_argument(
"--diar_input_size",
type=int_or_none,
default=None,
help="The number of input dimension of the feature",
)
group.add_argument(
"--enh_model_conf",
action=NestedDictAction,
default=get_default_kwargs(ESPnetEnhancementModel),
help="The keyword arguments for enh submodel class.",
)
group.add_argument(
"--asr_model_conf",
action=NestedDictAction,
default=get_default_kwargs(ESPnetASRModel),
help="The keyword arguments for asr submodel class.",
)
group.add_argument(
"--st_model_conf",
action=NestedDictAction,
default=get_default_kwargs(ESPnetEnhancementModel),
help="The keyword arguments for st submodel class.",
)
group.add_argument(
"--diar_model_conf",
action=NestedDictAction,
default=get_default_kwargs(ESPnetDiarizationModel),
help="The keyword arguments for diar submodel class.",
)
group.add_argument(
"--subtask_series",
type=str,
nargs="+",
default=("enh", "asr"),
choices=["enh", "asr", "st", "diar"],
help="The series of subtasks in the pipeline.",
)
group.add_argument(
"--model_conf",
action=NestedDictAction,
default=get_default_kwargs(ESPnetEnhS2TModel),
help="The keyword arguments for model class.",
)
group = parser.add_argument_group(description="Preprocess related")
group.add_argument(
"--use_preprocessor",
type=str2bool,
default=False,
help="Apply preprocessing to data or not",
)
group.add_argument(
"--token_type",
type=str,
default="bpe",
choices=["bpe", "char", "word", "phn"],
help="The text will be tokenized " "in the specified level token",
)
group.add_argument(
"--bpemodel",
type=str_or_none,
default=None,
help="The model file of sentencepiece",
)
group.add_argument(
"--src_token_type",
type=str,
default="bpe",
choices=["bpe", "char", "word", "phn"],
help="The source text will be tokenized " "in the specified level token",
)
group.add_argument(
"--src_bpemodel",
type=str_or_none,
default=None,
help="The model file of sentencepiece (for source language)",
)
group.add_argument(
"--non_linguistic_symbols",
type=str_or_none,
help="non_linguistic_symbols file path",
)
group.add_argument(
"--cleaner",
type=str_or_none,
choices=[None, "tacotron", "jaconv", "vietnamese"],
default=None,
help="Apply text cleaning",
)
group.add_argument(
"--g2p",
type=str_or_none,
choices=g2p_choices,
default=None,
help="Specify g2p method if --token_type=phn",
)
for class_choices in cls.class_choices_list:
# Append --<name> and --<name>_conf.
# e.g. --encoder and --encoder_conf
class_choices.add_arguments(group)
def add_task_arguments(cls, parser: argparse.ArgumentParser):
group = parser.add_argument_group(description="Task related")
required = parser.get_default("required")
required += ["token_list"]
group.add_argument("--train_pseudo_shape_file",
type=str,
action="append",
default=[])
group.add_argument(
"--train_pseudo_data_path_and_name_and_type",
type=str2triple_str,
action="append",
default=[],
help=
"Give three words splitted by comma. It's used for the training data.",
)
group.add_argument(
"--token_list",
type=str_or_none,
default=None,
help="A text mapping int-id to token",
)
group.add_argument(
"--pis_ratio",
type=float,
default=1,
help="The multi-iteration ratio for SSL",
)
group.add_argument(
"--init",
type=lambda x: str_or_none(x.lower()),
default=None,
help="The initialization method",
choices=[
"chainer",
"xavier_uniform",
"xavier_normal",
"kaiming_uniform",
"kaiming_normal",
None,
],
)
group.add_argument(
"--input_size",
type=int_or_none,
default=None,
help="The number of input dimension of the feature",
)
group.add_argument(
"--ctc_conf",
action=NestedDictAction,
default=get_default_kwargs(CTC),
help="The keyword arguments for CTC class.",
)
group.add_argument(
"--model_conf",
action=NestedDictAction,
default=get_default_kwargs(ESPnetASRModel),
help="The keyword arguments for model class.",
)
group = parser.add_argument_group(description="Preprocess related")
group.add_argument(
"--use_preprocessor",
type=str2bool,
default=True,
help="Apply preprocessing to data or not",
)
group.add_argument(
"--token_type",
type=str,
default="bpe",
choices=["bpe", "char", "word", "phn"],
help="The text will be tokenized "
"in the specified level token",
)
group.add_argument(
"--bpemodel",
type=str_or_none,
default=None,
help="The model file of sentencepiece",
)
parser.add_argument(
"--non_linguistic_symbols",
type=str_or_none,
help="non_linguistic_symbols file path",
)
parser.add_argument(
"--cleaner",
type=str_or_none,
choices=[None, "tacotron", "jaconv", "vietnamese"],
default=None,
help="Apply text cleaning",
)
parser.add_argument(
"--g2p",
type=str_or_none,
choices=[None, "g2p_en", "pyopenjtalk", "pyopenjtalk_kana"],
default=None,
help="Specify g2p method if --token_type=phn",
)
parser.add_argument(
"--speech_volume_normalize",
type=float_or_none,
default=None,
help="Scale the maximum amplitude to the given value.",
)
parser.add_argument(
"--rir_scp",
type=str_or_none,
default=None,
help="The file path of rir scp file.",
)
parser.add_argument(
"--rir_apply_prob",
type=float,
default=1.0,
help="THe probability for applying RIR convolution.",
)
parser.add_argument(
"--noise_scp",
type=str_or_none,
default=None,
help="The file path of noise scp file.",
)
parser.add_argument(
"--noise_apply_prob",
type=float,
default=1.0,
help="The probability applying Noise adding.",
)
parser.add_argument(
"--noise_db_range",
type=str,
default="13_15",
help="The range of noise decibel level.",
)
for class_choices in cls.class_choices_list:
# Append --<name> and --<name>_conf.
# e.g. --encoder and --encoder_conf
class_choices.add_arguments(group)
def __init__(
self,
vocab_size: int,
encoder_output_size: int,
rnn_type: str = "lstm",
num_layers: int = 1,
hidden_size: int = 320,
sampling_probability: float = 0.0,
dropout: float = 0.0,
context_residual: bool = False,
replace_sos: bool = False,
num_encs: int = 1,
att_conf: dict = get_default_kwargs(build_attention_list),
):
# FIXME(kamo): The parts of num_spk should be refactored more more more
assert check_argument_types()
if rnn_type not in {"lstm", "gru"}:
raise ValueError(f"Not supported: rnn_type={rnn_type}")
super().__init__()
eprojs = encoder_output_size
self.encoder_output_size = eprojs
self.dtype = rnn_type
self.dunits = hidden_size
self.dlayers = num_layers
self.context_residual = context_residual
self.sos = vocab_size - 1
self.eos = vocab_size - 1
self.odim = vocab_size
self.sampling_probability = sampling_probability
self.dropout = dropout
self.num_encs = num_encs
# for multilingual translation
self.replace_sos = replace_sos
self.embed = torch.nn.Embedding(vocab_size, hidden_size)
self.dropout_emb = torch.nn.Dropout(p=dropout)
self.decoder = torch.nn.ModuleList()
self.dropout_dec = torch.nn.ModuleList()
self.decoder += [
torch.nn.LSTMCell(hidden_size + eprojs, hidden_size) if self.dtype
== "lstm" else torch.nn.GRUCell(hidden_size + eprojs, hidden_size)
]
self.dropout_dec += [torch.nn.Dropout(p=dropout)]
for _ in range(1, self.dlayers):
self.decoder += [
torch.nn.LSTMCell(hidden_size, hidden_size) if self.dtype
== "lstm" else torch.nn.GRUCell(hidden_size, hidden_size)
]
self.dropout_dec += [torch.nn.Dropout(p=dropout)]
# NOTE: dropout is applied only for the vertical connections
# see https://arxiv.org/pdf/1409.2329.pdf
if context_residual:
self.output = torch.nn.Linear(hidden_size + eprojs, vocab_size)
else:
self.output = torch.nn.Linear(hidden_size, vocab_size)
self.att_list = build_attention_list(eprojs=eprojs,
dunits=hidden_size,
**att_conf)
def get_parser():
"""Get argument parser."""
parser = config_argparse.ArgumentParser(
description="TTS Decode",
formatter_class=argparse.ArgumentDefaultsHelpFormatter,
)
# Note(kamo): Use "_" instead of "-" as separator.
# "-" is confusing if written in yaml.
parser.add_argument(
"--log_level",
type=lambda x: x.upper(),
default="INFO",
choices=("CRITICAL", "ERROR", "WARNING", "INFO", "DEBUG", "NOTSET"),
help="The verbose level of logging",
)
parser.add_argument(
"--output_dir",
type=str,
required=True,
help="The path of output directory",
)
parser.add_argument(
"--ngpu",
type=int,
default=0,
help="The number of gpus. 0 indicates CPU mode",
)
parser.add_argument(
"--seed",
type=int,
default=0,
help="Random seed",
)
parser.add_argument(
"--dtype",
default="float32",
choices=["float16", "float32", "float64"],
help="Data type",
)
parser.add_argument(
"--num_workers",
type=int,
default=1,
help="The number of workers used for DataLoader",
)
parser.add_argument(
"--batch_size",
type=int,
default=1,
help="The batch size for inference",
)
group = parser.add_argument_group("Input data related")
group.add_argument(
"--data_path_and_name_and_type",
type=str2triple_str,
required=True,
action="append",
)
group.add_argument(
"--key_file",
type=str_or_none,
)
group.add_argument(
"--allow_variable_data_keys",
type=str2bool,
default=False,
)
group = parser.add_argument_group("The model configuration related")
group.add_argument(
"--train_config",
type=str,
help="Training configuration file.",
)
group.add_argument(
"--model_file",
type=str,
help="Model parameter file.",
)
group = parser.add_argument_group("Decoding related")
group.add_argument(
"--maxlenratio",
type=float,
default=10.0,
help="Maximum length ratio in decoding",
)
group.add_argument(
"--minlenratio",
type=float,
default=0.0,
help="Minimum length ratio in decoding",
)
group.add_argument(
"--threshold",
type=float,
default=0.5,
help="Threshold value in decoding",
)
group.add_argument(
"--use_att_constraint",
type=str2bool,
default=False,
help="Whether to use attention constraint",
)
group.add_argument(
"--backward_window",
type=int,
default=1,
help="Backward window value in attention constraint",
)
group.add_argument(
"--forward_window",
type=int,
default=3,
help="Forward window value in attention constraint",
)
group.add_argument(
"--use_teacher_forcing",
type=str2bool,
default=False,
help="Whether to use teacher forcing",
)
parser.add_argument(
"--speed_control_alpha",
type=float,
default=1.0,
help="Alpha in FastSpeech to change the speed of generated speech",
)
group = parser.add_argument_group("Grriffin-Lim related")
group.add_argument(
"--vocoder_conf",
action=NestedDictAction,
default=get_default_kwargs(Spectrogram2Waveform),
help="The configuration for Grriffin-Lim",
)
return parser
def add_task_arguments(cls, parser: argparse.ArgumentParser):
group = parser.add_argument_group(description="Task related")
# NOTE(kamo): add_arguments(..., required=True) can't be used
# to provide --print_config mode. Instead of it, do as
required = parser.get_default("required")
required += ["src_token_list", "token_list"]
group.add_argument(
"--token_list",
type=str_or_none,
default=None,
help="A text mapping int-id to token (for target language)",
)
group.add_argument(
"--src_token_list",
type=str_or_none,
default=None,
help="A text mapping int-id to token (for source language)",
)
group.add_argument(
"--init",
type=lambda x: str_or_none(x.lower()),
default=None,
help="The initialization method",
choices=[
"chainer",
"xavier_uniform",
"xavier_normal",
"kaiming_uniform",
"kaiming_normal",
None,
],
)
group.add_argument(
"--input_size",
type=int_or_none,
default=None,
help="The number of input dimension of the feature",
)
group.add_argument(
"--ctc_conf",
action=NestedDictAction,
default=get_default_kwargs(CTC),
help="The keyword arguments for CTC class.",
)
group.add_argument(
"--model_conf",
action=NestedDictAction,
default=get_default_kwargs(ESPnetSTModel),
help="The keyword arguments for model class.",
)
group = parser.add_argument_group(description="Preprocess related")
group.add_argument(
"--use_preprocessor",
type=str2bool,
default=True,
help="Apply preprocessing to data or not",
)
group.add_argument(
"--token_type",
type=str,
default="bpe",
choices=["bpe", "char", "word", "phn"],
help="The target text will be tokenized "
"in the specified level token",
)
group.add_argument(
"--src_token_type",
type=str,
default="bpe",
choices=["bpe", "char", "word", "phn"],
help="The source text will be tokenized "
"in the specified level token",
)
group.add_argument(
"--bpemodel",
type=str_or_none,
default=None,
help="The model file of sentencepiece (for target language)",
)
group.add_argument(
"--src_bpemodel",
type=str_or_none,
default=None,
help="The model file of sentencepiece (for source language)",
)
parser.add_argument(
"--non_linguistic_symbols",
type=str_or_none,
help="non_linguistic_symbols file path",
)
parser.add_argument(
"--cleaner",
type=str_or_none,
choices=[None, "tacotron", "jaconv", "vietnamese"],
default=None,
help="Apply text cleaning",
)
parser.add_argument(
"--g2p",
type=str_or_none,
choices=g2p_choices,
default=None,
help="Specify g2p method if --token_type=phn",
)
parser.add_argument(
"--speech_volume_normalize",
type=float_or_none,
default=None,
help="Scale the maximum amplitude to the given value.",
)
parser.add_argument(
"--rir_scp",
type=str_or_none,
default=None,
help="The file path of rir scp file.",
)
parser.add_argument(
"--rir_apply_prob",
type=float,
default=1.0,
help="THe probability for applying RIR convolution.",
)
parser.add_argument(
"--noise_scp",
type=str_or_none,
default=None,
help="The file path of noise scp file.",
)
parser.add_argument(
"--noise_apply_prob",
type=float,
default=1.0,
help="The probability applying Noise adding.",
)
parser.add_argument(
"--noise_db_range",
type=str,
default="13_15",
help="The range of noise decibel level.",
)
for class_choices in cls.class_choices_list:
# Append --<name> and --<name>_conf.
# e.g. --encoder and --encoder_conf
class_choices.add_arguments(group)
def add_task_arguments(cls, parser: argparse.ArgumentParser):
# NOTE(kamo): Use '_' instead of '-' to avoid confusion
assert check_argument_types()
group = parser.add_argument_group(description="Task related")
# NOTE(kamo): add_arguments(..., required=True) can't be used
# to provide --print_config mode. Instead of it, do as
required = parser.get_default("required")
required += ["token_list"]
group.add_argument(
"--token_list",
type=str_or_none,
default=None,
help="A train mapping int-id to token",
)
group.add_argument(
"--odim",
type=int_or_none,
default=None,
help="The number of dimension of output feature",
)
group.add_argument(
"--model_conf",
action=NestedDictAction,
default=get_default_kwargs(ESPnetTTSModel),
help="The keyword arguments for model class.",
)
group = parser.add_argument_group(description="Preprocess related")
group.add_argument(
"--use_preprocessor",
type=str2bool,
default=True,
help="Apply preprocessing to data or not",
)
group.add_argument(
"--token_type",
type=str,
default="phn",
choices=["bpe", "char", "word", "phn"],
help="The train will be tokenized in the specified level token",
)
group.add_argument(
"--bpemodel",
type=str_or_none,
default=None,
help="The model file of sentencepiece",
)
parser.add_argument(
"--non_linguistic_symbols",
type=str_or_none,
help="non_linguistic_symbols file path",
)
parser.add_argument(
"--cleaner",
type=str_or_none,
choices=[None, "tacotron", "jaconv", "vietnamese"],
default=None,
help="Apply train cleaning",
)
parser.add_argument(
"--g2p",
type=str_or_none,
choices=[
None,
"g2p_en",
"g2p_en_no_space",
"pyopenjtalk",
"pyopenjtalk_kana",
"pyopenjtalk_accent",
"pyopenjtalk_accent_with_pause",
"pypinyin_g2p",
"pypinyin_g2p_phone",
"espeak_ng_arabic",
],
default=None,
help="Specify g2p method if --token_type=phn",
)
for class_choices in cls.class_choices_list:
# Append --<name> and --<name>_conf.
# e.g. --encoder and --encoder_conf
class_choices.add_arguments(group)
def test_get_defaut_kwargs(func, desired: Any):
assert get_default_kwargs(func) == desired
def add_task_arguments(cls, parser: argparse.ArgumentParser):
group = parser.add_argument_group(description="Task related")
# NOTE(kamo): add_arguments(..., required=True) can't be used
# to provide --print_config mode. Instead of it, do as
required = parser.get_default("required")
group.add_argument(
"--model_conf",
action=NestedDictAction,
default=get_default_kwargs(CepNet),
help="The keyword arguments for model class.",
)
group.add_argument(
"--init",
type=lambda x: str_or_none(x.lower()),
default=None,
help="The initialization method",
choices=[
"chainer",
"xavier_uniform",
"xavier_normal",
"kaiming_uniform",
"kaiming_normal",
None,
],
)
group.add_argument(
"--input_size",
type=int_or_none,
default=None,
help="The number of input dimension of the feature",
)
group = parser.add_argument_group(description="Preprocess related")
group.add_argument(
"--use_preprocessor",
type=str2bool,
default=True,
help="Apply preprocessing to data or not",
)
parser.add_argument(
"--speech_volume_normalize",
type=float_or_none,
default=None,
help="Scale the maximum amplitude to the given value.",
)
parser.add_argument(
"--rir_scp",
type=str_or_none,
default=None,
help="The file path of rir scp file.",
)
parser.add_argument(
"--rir_apply_prob",
type=float,
default=1.0,
help="THe probability for applying RIR convolution.",
)
parser.add_argument(
"--noise_scp",
type=str_or_none,
default=None,
help="The file path of noise scp file.",
)
parser.add_argument(
"--noise_apply_prob",
type=float,
default=1.0,
help="The probability applying Noise adding.",
)
parser.add_argument(
"--noise_db_range",
type=str,
default="13_15",
help="The range of noise decibel level.",
)
for class_choices in cls.class_choices_list:
# Append --<name> and --<name>_conf.
# e.g. --encoder and --encoder_conf
class_choices.add_arguments(group)
def add_task_arguments(cls, parser: argparse.ArgumentParser):
# NOTE(kamo): Use '_' instead of '-' to avoid confusion
assert check_argument_types()
group = parser.add_argument_group(description="Task related")
# NOTE(kamo): add_arguments(..., required=True) can't be used
# to provide --print_config mode. Instead of it, do as
required = parser.get_default("required")
required += ["token_list"]
group.add_argument(
"--token_list",
type=str_or_none,
default=None,
help="A text mapping int-id to token",
)
group.add_argument(
"--init",
type=lambda x: str_or_none(x.lower()),
default=None,
help="The initialization method",
choices=[
"chainer",
"xavier_uniform",
"xavier_normal",
"kaiming_uniform",
"kaiming_normal",
None,
],
)
group.add_argument(
"--model_conf",
action=NestedDictAction,
default=get_default_kwargs(ESPnetLanguageModel),
help="The keyword arguments for model class.",
)
group = parser.add_argument_group(description="Preprocess related")
group.add_argument(
"--use_preprocessor",
type=str2bool,
default=True,
help="Apply preprocessing to data or not",
)
group.add_argument(
"--token_type",
type=str,
default="bpe",
choices=["bpe", "char", "word"],
help="",
)
group.add_argument(
"--bpemodel",
type=str_or_none,
default=None,
help="The model file fo sentencepiece",
)
parser.add_argument(
"--non_linguistic_symbols",
type=str_or_none,
help="non_linguistic_symbols file path",
)
parser.add_argument(
"--cleaner",
type=str_or_none,
choices=[None, "tacotron", "jaconv", "vietnamese"],
default=None,
help="Apply text cleaning",
)
parser.add_argument(
"--g2p",
type=str_or_none,
choices=[
None,
"g2p_en",
"g2p_en_no_space",
"pyopenjtalk",
"pyopenjtalk_kana",
"pyopenjtalk_accent",
"pyopenjtalk_accent_with_pause",
"pypinyin_g2p",
"pypinyin_g2p_phone",
"espeak_ng_arabic",
"espeak_ng_german",
"espeak_ng_french",
"espeak_ng_spanish",
"espeak_ng_russian",
"g2pk",
"g2pk_no_space",
],
default=None,
help="Specify g2p method if --token_type=phn",
)
for class_choices in cls.class_choices_list:
# Append --<name> and --<name>_conf.
# e.g. --encoder and --encoder_conf
class_choices.add_arguments(group)
assert check_return_type(parser)
return parser
def add_task_arguments(cls, parser: argparse.ArgumentParser):
group = parser.add_argument_group(description="Task related")
# NOTE(kamo): add_arguments(..., required=True) can't be used
# to provide --print_config mode. Instead of it, do as
required = parser.get_default("required")
required += ["token_list"]
group.add_argument(
"--token_list",
type=str_or_none,
default=None,
help="A text mapping int-id to token",
)
group.add_argument(
"--init",
type=lambda x: str_or_none(x.lower()),
default=None,
help="The initialization method",
choices=[
"chainer",
"xavier_uniform",
"xavier_normal",
"kaiming_uniform",
"kaiming_normal",
None,
],
)
group.add_argument(
"--input_size",
type=int_or_none,
default=None,
help="The number of input dimension of the feature",
)
group.add_argument(
"--ctc_conf",
action=NestedDictAction,
default=get_default_kwargs(CTC),
help="The keyword arguments for CTC class.",
)
group.add_argument(
"--asr_model_conf",
action=NestedDictAction,
default=get_default_kwargs(ESPnetASRModel),
help="The keyword arguments for model class.",
)
group.add_argument(
"--enh_model_conf",
action=NestedDictAction,
default=get_default_kwargs(ESPnetEnhancementModel),
help="The keyword arguments for model class.",
)
group = parser.add_argument_group(description="Preprocess related")
group.add_argument(
"--use_preprocessor",
type=str2bool,
default=False,
help="Apply preprocessing to data or not",
)
group.add_argument(
"--token_type",
type=str,
default="bpe",
choices=["bpe", "char", "word", "phn"],
help="The text will be tokenized " "in the specified level token",
)
group.add_argument(
"--bpemodel",
type=str_or_none,
default=None,
help="The model file of sentencepiece",
)
parser.add_argument(
"--non_linguistic_symbols",
type=str_or_none,
help="non_linguistic_symbols file path",
)
parser.add_argument(
"--cleaner",
type=str_or_none,
choices=[None, "tacotron", "jaconv", "vietnamese"],
default=None,
help="Apply text cleaning",
)
parser.add_argument(
"--g2p",
type=str_or_none,
choices=[None, "g2p_en", "pyopenjtalk", "pyopenjtalk_kana"],
default=None,
help="Specify g2p method if --token_type=phn",
)
for class_choices in cls.class_choices_list:
# Append --<name> and --<name>_conf.
# e.g. --encoder and --encoder_conf
class_choices.add_arguments(group)
def add_task_arguments(cls, parser: argparse.ArgumentParser):
group = parser.add_argument_group(description="Task related")
# NOTE(kamo): add_arguments(..., required=True) can't be used
# to provide --print_config mode. Instead of it, do as
# required = parser.get_default("required")
group.add_argument(
"--init",
type=lambda x: str_or_none(x.lower()),
default=None,
help="The initialization method",
choices=[
"chainer",
"xavier_uniform",
"xavier_normal",
"kaiming_uniform",
"kaiming_normal",
None,
],
)
group.add_argument(
"--model_conf",
action=NestedDictAction,
default=get_default_kwargs(ESPnetEnhancementModel),
help="The keyword arguments for model class.",
)
group.add_argument(
"--criterions",
action=NestedDictAction,
default=[
{
"name": "si_snr",
"conf": {},
"wrapper": "fixed_order",
"wrapper_conf": {},
},
],
help="The criterions binded with the loss wrappers.",
)
group = parser.add_argument_group(description="Preprocess related")
group.add_argument(
"--use_preprocessor",
type=str2bool,
default=False,
help="Whether to apply preprocessing to data or not",
)
group.add_argument(
"--speech_volume_normalize",
type=str_or_none,
default=None,
help="Scale the maximum amplitude to the given value or range. "
"e.g. --speech_volume_normalize 1.0 scales it to 1.0.\n"
"--speech_volume_normalize 0.5_1.0 scales it to a random number in "
"the range [0.5, 1.0)",
)
group.add_argument(
"--rir_scp",
type=str_or_none,
default=None,
help="The file path of rir scp file.",
)
group.add_argument(
"--rir_apply_prob",
type=float,
default=1.0,
help="THe probability for applying RIR convolution.",
)
group.add_argument(
"--noise_scp",
type=str_or_none,
default=None,
help="The file path of noise scp file.",
)
group.add_argument(
"--noise_apply_prob",
type=float,
default=1.0,
help="The probability applying Noise adding.",
)
group.add_argument(
"--noise_db_range",
type=str,
default="13_15",
help="The range of signal-to-noise ratio (SNR) level in decibel.",
)
group.add_argument(
"--short_noise_thres",
type=float,
default=0.5,
help=
"If len(noise) / len(speech) is smaller than this threshold during "
"dynamic mixing, a warning will be displayed.",
)
group.add_argument(
"--use_reverberant_ref",
type=str2bool,
default=False,
help="Whether to use reverberant speech references "
"instead of anechoic ones",
)
group.add_argument(
"--num_spk",
type=int,
default=1,
help="Number of speakers in the input signal.",
)
group.add_argument(
"--num_noise_type",
type=int,
default=1,
help="Number of noise types.",
)
group.add_argument(
"--sample_rate",
type=int,
default=8000,
help="Sampling rate of the data (in Hz).",
)
group.add_argument(
"--force_single_channel",
type=str2bool,
default=False,
help="Whether to force all data to be single-channel.",
)
for class_choices in cls.class_choices_list:
# Append --<name> and --<name>_conf.
# e.g. --encoder and --encoder_conf
class_choices.add_arguments(group)
def add_task_arguments(cls, parser: argparse.ArgumentParser):
# NOTE(kamo): Use '_' instead of '-' to avoid confusion
assert check_argument_types()
group = parser.add_argument_group(description="Task related")
# NOTE(kamo): add_arguments(..., required=True) can't be used
# to provide --print_config mode. Instead of it, do as
required = parser.get_default("required")
required += ["token_list"]
group.add_argument(
"--token_list",
type=str_or_none,
default=None,
help="A text mapping int-id to token",
)
group.add_argument(
"--init",
type=lambda x: str_or_none(x.lower()),
default=None,
help="The initialization method",
choices=[
"chainer",
"xavier_uniform",
"xavier_normal",
"kaiming_uniform",
"kaiming_normal",
None,
],
)
group.add_argument(
"--model_conf",
action=NestedDictAction,
default=get_default_kwargs(ESPnetLanguageModel),
help="The keyword arguments for model class.",
)
group = parser.add_argument_group(description="Preprocess related")
group.add_argument(
"--use_preprocessor",
type=str2bool,
default=False,
help="Apply preprocessing to data or not",
)
group.add_argument(
"--token_type",
type=str,
default="bpe",
choices=["bpe", "char", "word"],
help="",
)
group.add_argument(
"--bpemodel",
type=str_or_none,
default=None,
help="The model file fo sentencepiece",
)
parser.add_argument(
"--non_linguistic_symbols",
type=str_or_none,
help="non_linguistic_symbols file path",
)
for class_choices in cls.class_choices_list:
# Append --<name> and --<name>_conf.
# e.g. --encoder and --encoder_conf
class_choices.add_arguments(group)
assert check_return_type(parser)
return parser
def add_task_arguments(cls, parser: argparse.ArgumentParser):
"""Add Transducer task arguments.
Args:
cls: ASRTransducerTask object.
parser: Transducer arguments parser.
"""
group = parser.add_argument_group(description="Task related")
required = parser.get_default("required")
required += ["token_list"]
group.add_argument(
"--token_list",
type=str_or_none,
default=None,
help="A text mapping int-id to token",
)
group.add_argument(
"--input_size",
type=int_or_none,
default=None,
help="The number of input dimension of the feature",
)
group.add_argument(
"--init",
type=lambda x: str_or_none(x.lower()),
default=None,
help="The initialization method",
choices=[
"chainer",
"chainer_espnet1",
"xavier_uniform",
"xavier_normal",
"kaiming_uniform",
"kaiming_normal",
None,
],
)
group.add_argument(
"--model_conf",
action=NestedDictAction,
default=get_default_kwargs(ESPnetASRTransducerModel),
help="The keyword arguments for model class.",
)
group.add_argument(
"--encoder_conf",
action=NestedDictAction,
default={},
help="The keyword arguments for encoder class.",
)
group.add_argument(
"--joint_network_conf",
action=NestedDictAction,
default={},
help="The keyword arguments for joint network class.",
)
group = parser.add_argument_group(description="Preprocess related")
group.add_argument(
"--use_preprocessor",
type=str2bool,
default=True,
help="Apply preprocessing to data or not",
)
group.add_argument(
"--token_type",
type=str,
default="bpe",
choices=["bpe", "char", "word", "phn"],
help="The text will be tokenized "
"in the specified level token",
)
group.add_argument(
"--bpemodel",
type=str_or_none,
default=None,
help="The model file of sentencepiece",
)
parser.add_argument(
"--non_linguistic_symbols",
type=str_or_none,
help="non_linguistic_symbols file path",
)
parser.add_argument(
"--cleaner",
type=str_or_none,
choices=[None, "tacotron", "jaconv", "vietnamese"],
default=None,
help="Apply text cleaning",
)
parser.add_argument(
"--g2p",
type=str_or_none,
choices=g2p_choices,
default=None,
help="Specify g2p method if --token_type=phn",
)
parser.add_argument(
"--speech_volume_normalize",
type=float_or_none,
default=None,
help="Scale the maximum amplitude to the given value.",
)
parser.add_argument(
"--rir_scp",
type=str_or_none,
default=None,
help="The file path of rir scp file.",
)
parser.add_argument(
"--rir_apply_prob",
type=float,
default=1.0,
help="The probability for applying RIR convolution.",
)
parser.add_argument(
"--noise_scp",
type=str_or_none,
default=None,
help="The file path of noise scp file.",
)
parser.add_argument(
"--noise_apply_prob",
type=float,
default=1.0,
help="The probability applying Noise adding.",
)
parser.add_argument(
"--noise_db_range",
type=str,
default="13_15",
help="The range of noise decibel level.",
)
for class_choices in cls.class_choices_list:
# Append --<name> and --<name>_conf.
# e.g. --decoder and --decoder_conf
class_choices.add_arguments(group)
