def __init__(
self,
wtype: str = "blstmp",
widim: int = 257,
wlayers: int = 3,
wunits: int = 300,
wprojs: int = 320,
dropout_rate: float = 0.0,
taps: int = 5,
delay: int = 3,
use_dnn_mask: bool = True,
nmask: int = 1,
nonlinear: str = "sigmoid",
iterations: int = 1,
normalization: bool = False,
eps: float = 1e-6,
diagonal_loading: bool = True,
diag_eps: float = 1e-7,
mask_flooring: bool = False,
flooring_thres: float = 1e-6,
use_torch_solver: bool = True,
):
super().__init__()
self.iterations = iterations
self.taps = taps
self.delay = delay
self.eps = eps
self.normalization = normalization
self.use_dnn_mask = use_dnn_mask
self.inverse_power = True
self.diagonal_loading = diagonal_loading
self.diag_eps = diag_eps
self.mask_flooring = mask_flooring
self.flooring_thres = flooring_thres
self.use_torch_solver = use_torch_solver
if self.use_dnn_mask:
self.nmask = nmask
self.mask_est = MaskEstimator(
wtype,
widim,
wlayers,
wunits,
wprojs,
dropout_rate,
nmask=nmask,
nonlinear=nonlinear,
)
else:
self.nmask = 1
def __init__(
self,
wtype: str = "blstmp",
widim: int = 257,
wlayers: int = 3,
wunits: int = 300,
wprojs: int = 320,
dropout_rate: float = 0.0,
taps: int = 5,
delay: int = 3,
use_dnn_mask: bool = True,
nonlinear: str = "sigmoid",
iterations: int = 1,
normalization: bool = False,
):
super().__init__()
self.iterations = iterations
self.taps = taps
self.delay = delay
self.normalization = normalization
self.use_dnn_mask = use_dnn_mask
self.inverse_power = True
if self.use_dnn_mask:
self.mask_est = MaskEstimator(
wtype,
widim,
wlayers,
wunits,
wprojs,
dropout_rate,
nmask=1,
nonlinear=nonlinear,
)
def __init__(
self,
bidim,
btype: str = "blstmp",
blayers: int = 3,
bunits: int = 300,
bprojs: int = 320,
num_spk: int = 1,
use_noise_mask: bool = True,
nonlinear: str = "sigmoid",
dropout_rate: float = 0.0,
badim: int = 320,
ref_channel: int = -1,
beamformer_type: str = "mvdr_souden",
rtf_iterations: int = 2,
mwf_mu: float = 1.0,
eps: float = 1e-6,
diagonal_loading: bool = True,
diag_eps: float = 1e-7,
mask_flooring: bool = False,
flooring_thres: float = 1e-6,
use_torch_solver: bool = True,
# only for WPD beamformer
btaps: int = 5,
bdelay: int = 3,
):
super().__init__()
bnmask = num_spk + 1 if use_noise_mask else num_spk
self.mask = MaskEstimator(
btype,
bidim,
blayers,
bunits,
bprojs,
dropout_rate,
nmask=bnmask,
nonlinear=nonlinear,
)
self.ref = (
AttentionReference(bidim, badim, eps=eps) if ref_channel < 0 else None
)
self.ref_channel = ref_channel
self.use_noise_mask = use_noise_mask
assert num_spk >= 1, num_spk
self.num_spk = num_spk
self.nmask = bnmask
if beamformer_type not in BEAMFORMER_TYPES:
raise ValueError("Not supporting beamformer_type=%s" % beamformer_type)
if (
beamformer_type == "mvdr_souden" or not beamformer_type.endswith("_souden")
) and not use_noise_mask:
if num_spk == 1:
logging.warning(
"Initializing %s beamformer without noise mask "
"estimator (single-speaker case)" % beamformer_type.upper()
)
logging.warning(
"(1 - speech_mask) will be used for estimating noise "
"PSD in %s beamformer!" % beamformer_type.upper()
)
else:
logging.warning(
"Initializing %s beamformer without noise mask "
"estimator (multi-speaker case)" % beamformer_type.upper()
)
logging.warning(
"Interference speech masks will be used for estimating "
"noise PSD in %s beamformer!" % beamformer_type.upper()
)
self.beamformer_type = beamformer_type
if not beamformer_type.endswith("_souden"):
assert rtf_iterations >= 2, rtf_iterations
# number of iterations in power method for estimating the RTF
self.rtf_iterations = rtf_iterations
# noise suppression weight in SDW-MWF
self.mwf_mu = mwf_mu
assert btaps >= 0 and bdelay >= 0, (btaps, bdelay)
self.btaps = btaps
self.bdelay = bdelay if self.btaps > 0 else 1
self.eps = eps
self.diagonal_loading = diagonal_loading
self.diag_eps = diag_eps
self.mask_flooring = mask_flooring
self.flooring_thres = flooring_thres
self.use_torch_solver = use_torch_solver
def __init__(
self,
bidim,
btype: str = "blstmp",
blayers: int = 3,
bunits: int = 300,
bprojs: int = 320,
num_spk: int = 1,
use_noise_mask: bool = True,
nonlinear: str = "sigmoid",
dropout_rate: float = 0.0,
badim: int = 320,
ref_channel: int = -1,
beamformer_type: str = "mvdr",
eps: float = 1e-6,
# only for WPD beamformer
btaps: int = 5,
bdelay: int = 3,
):
super().__init__()
bnmask = num_spk + 1 if use_noise_mask else num_spk
self.mask = MaskEstimator(
btype,
bidim,
blayers,
bunits,
bprojs,
dropout_rate,
nmask=bnmask,
nonlinear=nonlinear,
)
self.ref = AttentionReference(bidim,
badim) if ref_channel < 0 else None
self.ref_channel = ref_channel
self.use_noise_mask = use_noise_mask
assert num_spk >= 1, num_spk
self.num_spk = num_spk
self.nmask = bnmask
if beamformer_type not in ("mvdr", "mpdr", "wpd"):
raise ValueError(
"Not supporting beamformer_type={}".format(beamformer_type))
if beamformer_type == "mvdr" and (not use_noise_mask):
if num_spk == 1:
logging.warning(
"Initializing MVDR beamformer without noise mask "
"estimator (single-speaker case)")
logging.warning(
"(1 - speech_mask) will be used for estimating noise "
"PSD in MVDR beamformer!")
else:
logging.warning(
"Initializing MVDR beamformer without noise mask "
"estimator (multi-speaker case)")
logging.warning(
"Interference speech masks will be used for estimating "
"noise PSD in MVDR beamformer!")
self.beamformer_type = beamformer_type
assert btaps >= 0 and bdelay >= 0, (btaps, bdelay)
self.btaps = btaps
self.bdelay = bdelay if self.btaps > 0 else 1
self.eps = eps