def __init__(
self,
input_size: int,
rnn_type: str = "lstm",
bidirectional: bool = True,
use_projection: bool = True,
num_layers: int = 4,
hidden_size: int = 320,
output_size: int = 320,
dropout: float = 0.0,
in_channel: int = 1,
):
assert check_argument_types()
super().__init__()
self._output_size = output_size
self.rnn_type = rnn_type
self.bidirectional = bidirectional
self.use_projection = use_projection
if rnn_type not in {"lstm", "gru"}:
raise ValueError(f"Not supported rnn_type={rnn_type}")
# Subsample is not used for VGGRNN
subsample = np.ones(num_layers + 1, dtype=np.int64)
rnn_type = ("b" if bidirectional else "") + rnn_type
if use_projection:
self.enc = torch.nn.ModuleList([
VGG2L(in_channel),
RNNP(
get_vgg2l_odim(input_size, in_channel=in_channel),
num_layers,
hidden_size,
output_size,
subsample,
dropout,
typ=rnn_type,
),
])
else:
self.enc = torch.nn.ModuleList([
VGG2L(in_channel),
RNN(
get_vgg2l_odim(input_size, in_channel=in_channel),
num_layers,
hidden_size,
output_size,
dropout,
typ=rnn_type,
),
])
def __init__(
self,
etype,
idim,
elayers_sd,
elayers_rec,
eunits,
eprojs,
subsample,
dropout,
num_spkrs=2,
in_channel=1,
):
"""Initialize the encoder of single-channel multi-speaker ASR."""
super(EncoderMix, self).__init__()
typ = etype.lstrip("vgg").rstrip("p")
if typ not in ["lstm", "gru", "blstm", "bgru"]:
logging.error(
"Error: need to specify an appropriate encoder architecture")
if etype.startswith("vgg"):
if etype[-1] == "p":
self.enc_mix = torch.nn.ModuleList([VGG2L(in_channel)])
self.enc_sd = torch.nn.ModuleList([
torch.nn.ModuleList([
RNNP(
get_vgg2l_odim(idim, in_channel=in_channel),
elayers_sd,
eunits,
eprojs,
subsample[:elayers_sd + 1],
dropout,
typ=typ,
)
]) for i in range(num_spkrs)
])
self.enc_rec = torch.nn.ModuleList([
RNNP(
eprojs,
elayers_rec,
eunits,
eprojs,
subsample[elayers_sd:],
dropout,
typ=typ,
)
])
logging.info("Use CNN-VGG + B" + typ.upper() + "P for encoder")
else:
logging.error(
f"Error: need to specify an appropriate encoder architecture. "
f"Illegal name {etype}")
sys.exit()
else:
logging.error(
f"Error: need to specify an appropriate encoder architecture. "
f"Illegal name {etype}")
sys.exit()
self.num_spkrs = num_spkrs
def __init__(self,
etype,
idim,
elayers_sd,
elayers_rec,
eunits,
eprojs,
subsample,
dropout,
num_spkrs=2,
in_channel=1):
super(Encoder, self).__init__()
typ = etype.lstrip("vgg").lstrip("b").rstrip("p")
if typ != "lstm" and typ != "gru":
logging.error(
"Error: need to specify an appropriate encoder architecture")
if etype.startswith("vgg"):
if etype[-1] == "p":
self.enc_mix = torch.nn.ModuleList([VGG2L(in_channel)])
self.enc_sd = torch.nn.ModuleList([
torch.nn.ModuleList([
RNNP(get_vgg2l_odim(idim, in_channel=in_channel),
elayers_sd,
eunits,
eprojs,
subsample[:elayers_sd + 1],
dropout,
typ=typ)
]) for i in range(num_spkrs)
])
self.enc_rec = torch.nn.ModuleList([
RNNP(eprojs,
elayers_rec,
eunits,
eprojs,
subsample[elayers_sd:],
dropout,
typ=typ)
])
logging.info('Use CNN-VGG + B' + typ.upper() + 'P for encoder')
else:
logging.error(
"Error: need to specify an appropriate encoder architecture")
sys.exit()
self.num_spkrs = num_spkrs