from collections import OrderedDict
from typing import List
from typing import Tuple
from typing import Union
import torch
from torch_complex.tensor import ComplexTensor
from espnet2.enh.layers.complex_utils import is_complex
from espnet2.enh.layers.skim import SkiM
from espnet2.enh.separator.abs_separator import AbsSeparator
[docs]class SkiMSeparator(AbsSeparator):
"""Skipping Memory (SkiM) Separator
Args:
input_dim: input feature dimension
causal: bool, whether the system is causal.
num_spk: number of target speakers.
nonlinear: the nonlinear function for mask estimation,
select from 'relu', 'tanh', 'sigmoid'
layer: int, number of SkiM blocks. Default is 3.
unit: int, dimension of the hidden state.
segment_size: segmentation size for splitting long features
dropout: float, dropout ratio. Default is 0.
mem_type: 'hc', 'h', 'c', 'id' or None.
It controls whether the hidden (or cell) state of
SegLSTM will be processed by MemLSTM.
In 'id' mode, both the hidden and cell states
will be identically returned.
When mem_type is None, the MemLSTM will be removed.
seg_overlap: Bool, whether the segmentation will reserve 50%
overlap for adjacent segments. Default is False.
"""
def __init__(
self,
input_dim: int,
causal: bool = True,
num_spk: int = 2,
nonlinear: str = "relu",
layer: int = 3,
unit: int = 512,
segment_size: int = 20,
dropout: float = 0.0,
mem_type: str = "hc",
seg_overlap: bool = False,
):
super().__init__()
self._num_spk = num_spk
self.segment_size = segment_size
if mem_type not in ("hc", "h", "c", "id", None):
raise ValueError("Not supporting mem_type={}".format(mem_type))
self.skim = SkiM(
input_size=input_dim,
hidden_size=unit,
output_size=input_dim * num_spk,
dropout=dropout,
num_blocks=layer,
bidirectional=(not causal),
norm_type="cLN" if causal else "gLN",
segment_size=segment_size,
seg_overlap=seg_overlap,
mem_type=mem_type,
)
if nonlinear not in ("sigmoid", "relu", "tanh"):
raise ValueError("Not supporting nonlinear={}".format(nonlinear))
self.nonlinear = {
"sigmoid": torch.nn.Sigmoid(),
"relu": torch.nn.ReLU(),
"tanh": torch.nn.Tanh(),
}[nonlinear]
[docs] def forward(
self, input: Union[torch.Tensor, ComplexTensor], ilens: torch.Tensor
) -> Tuple[List[Union[torch.Tensor, ComplexTensor]], torch.Tensor, OrderedDict]:
"""Forward.
Args:
input (torch.Tensor or ComplexTensor): Encoded feature [B, T, N]
ilens (torch.Tensor): input lengths [Batch]
Returns:
masked (List[Union(torch.Tensor, ComplexTensor)]): [(B, T, N), ...]
ilens (torch.Tensor): (B,)
others predicted data, e.g. masks: OrderedDict[
'mask_spk1': torch.Tensor(Batch, Frames, Freq),
'mask_spk2': torch.Tensor(Batch, Frames, Freq),
...
'mask_spkn': torch.Tensor(Batch, Frames, Freq),
]
"""
# if complex spectrum,
if is_complex(input):
feature = abs(input)
else:
feature = input
B, T, N = feature.shape
processed = self.skim(feature) # B,T, N
processed = processed.view(B, T, N, self.num_spk)
masks = self.nonlinear(processed).unbind(dim=3)
masked = [input * m for m in masks]
others = OrderedDict(
zip(["mask_spk{}".format(i + 1) for i in range(len(masks))], masks)
)
return masked, ilens, others
@property
def num_spk(self):
return self._num_spk