o i@stddlmZddlmZmZmZmZmZddlZddl m Z ddl m Z ddl mZddlmZGdd d eZdS) ) OrderedDict)DictListOptionalTupleUnionN) ComplexTensor) is_complex)SkiM) AbsSeparatorcseZdZdZ        d!d ed ed ededededededededeffdd Z d"de e j e fde j de edeee e j e fe j effddZedd ZZS)# SkiMSeparatoraSkipping 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. TFreluhc input_dimcausalnum_spk predict_noise nonlinearlayerunit segment_sizedropoutmem_type seg_overlapc st||_||_||_| dvrtd| |jr!|jdn|j|_t ||||j| || |r4dnd|| | d |_ |dvrGtd|t j t j t j d||_dS) N)rhcidNzNot supporting mem_type={}cLNgLN) input_size hidden_size output_sizer num_blocks bidirectional norm_typerrr)sigmoidrtanhzNot supporting nonlinear={})super__init___num_spkrr ValueErrorformatr num_outputsr skimtorchnnSigmoidReLUTanhr) selfrrrrrrrrrrr __class__X/home/ubuntu/.local/lib/python3.10/site-packages/espnet2/enh/separator/skim_separator.pyr.#s6    zSkiMSeparator.__init__Ninputilens additionalreturnc str t}n}|j\}}}||}|||||j}||jdd} |jr0| ^} } fdd| D} t t ddt t | D| } |jrR| | d<| || fS)aForward. Args: input (torch.Tensor or ComplexTensor): Encoded feature [B, T, N] ilens (torch.Tensor): input lengths [Batch] additional (Dict or None): other data included in model NOTE: not used in this model 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), ] r)dimcsg|]}|qSr<r<).0mr>r<r= {sz)SkiMSeparator.forward..cSsg|] }d|dqS)z mask_spk{}r")r1)rCir<r<r=rF~snoise1) r absshaper3viewr2runbindrrziprangelen) r9r>r?r@featureBTN processedmasks mask_noisemaskedothersr<rEr=forwardSs       zSkiMSeparator.forwardcCs|jSN)r/)r9r<r<r=rszSkiMSeparator.num_spk) Tr FrrrrrrFrZ)__name__ __module__ __qualname____doc__intboolstrfloatr.rr4TensorrrrrrrrYpropertyr __classcell__r<r<r:r=r s^    4  2r ) collectionsrtypingrrrrrr4torch_complex.tensorr espnet2.enh.layers.complex_utilsr espnet2.enh.layers.skimr #espnet2.enh.separator.abs_separatorr r r<r<r<r=s