o si"@sddlmZddlZddlmZddlmZddlmZm Z ddl Z ddl m Z ddl mZGd d d ejZGd d d ejZdS) )ceilN)MultiheadAttention) activationsnorms)has_arg)DualPathProcessingcs2eZdZdZ    d fdd Zdd ZZS) ImprovedTransformedLayera Improved Transformer module as used in [1]. It is Multi-Head self-attention followed by LSTM, activation and linear projection layer. Args: embed_dim (int): Number of input channels. n_heads (int): Number of attention heads. dim_ff (int): Number of neurons in the RNNs cell state. Defaults to 256. RNN here replaces standard FF linear layer in plain Transformer. dropout (float, optional): Dropout ratio, must be in [0,1]. activation (str, optional): activation function applied at the output of RNN. bidirectional (bool, optional): True for bidirectional Inter-Chunk RNN (Intra-Chunk is always bidirectional). norm (str, optional): Type of normalization to use. References [1] Chen, Jingjing, Qirong Mao, and Dong Liu. "Dual-Path Transformer Network: Direct Context-Aware Modeling for End-to-End Monaural Speech Separation." arXiv (2020). reluTgLNc stt|t|||d|_t||_tj|||dd|_ |r%d|n|}t |||_ t ||_t |||_t |||_dS)N)dropoutT) bidirectional batch_firstr)superr __init__rmhannDropoutrLSTM recurrentLinearlinearrget activationrnorm_mhanorm_ff) self embed_dimn_headsdim_ffrrrnorm ff_inner_dim __class__M/home/ubuntu/.local/lib/python3.10/site-packages/asteroid/masknn/attention.pyr"s  z!ImprovedTransformedLayer.__init__c Cs|ddd}||||d}||ddd|}||}|||||ddd}||dd|}||S)Nrrr) permuterrrrrr transposer)rxtomhaoutr&r&r'forward7s ( z ImprovedTransformedLayer.forward)r r Tr )__name__ __module__ __qualname____doc__rr. __classcell__r&r&r$r'r sr csFeZdZdZ         dfd d Zd dZddZZS) DPTransformeraDual-path Transformer introduced in [1]. Args: in_chan (int): Number of input filters. n_src (int): Number of masks to estimate. n_heads (int): Number of attention heads. ff_hid (int): Number of neurons in the RNNs cell state. Defaults to 256. chunk_size (int): window size of overlap and add processing. Defaults to 100. hop_size (int or None): hop size (stride) of overlap and add processing. Default to `chunk_size // 2` (50% overlap). n_repeats (int): Number of repeats. Defaults to 6. norm_type (str, optional): Type of normalization to use. ff_activation (str, optional): activation function applied at the output of RNN. mask_act (str, optional): Which non-linear function to generate mask. bidirectional (bool, optional): True for bidirectional Inter-Chunk RNN (Intra-Chunk is always bidirectional). dropout (float, optional): Dropout ratio, must be in [0,1]. References [1] Chen, Jingjing, Qirong Mao, and Dong Liu. "Dual-Path Transformer Network: Direct Context-Aware Modeling for End-to-End Monaural Speech Separation." arXiv (2020). dNr r Trc stt|||_||_||_||_||_|dur|n|d}||_||_ ||_||_ | |_ | |_ | |_ | |_t|j|j|j|_|j|jdkrltd|jd|jd|jd|jd t|j|j|_nd|_t||j|_t|j|j|_tg|_t|j D]*} |jtt|j|j|j|j|j d|j t|j|j|j|j|j |j |j gqt |j||jd }t!t"||_#t!t$|j|jd t%|_&t!t$|j|jd t'|_(t)| }t*|d r|d d |_+dS||_+dS) NrrzDPTransformer input dim (z,) is not a multiple of the number of heads (z;). Adding extra linear layer at input to accomodate (size [z x z])Trdim)r9),rr4rin_chann_srcr ff_hid chunk_sizehop_size n_repeats norm_type ff_activationmask_actrrr mha_in_dimwarningswarnrr input_layerrrin_normr ola ModuleListlayersrangeappendr Conv2d SequentialPReLU first_outConv1dTanhnet_outSigmoidnet_gaterr output_act)rr:r;r r<r=r>r?r@rArBrrr+ net_out_conv mask_nl_classr$r&r'ras~        zDPTransformer.__init__c Cs|jdur||dddd}||}|jd}|j|}|\}}|_}tt |j D]}|j |\}}|j ||}|j ||}q2| |} | ||j|j|j|} |jj| |d} || || } | ||j|jd} || } | S)zForward. Args: mixture_w (:class:`torch.Tensor`): Tensor of shape $(batch, nfilters, nframes)$ Returns: :class:`torch.Tensor`: estimated mask of shape $(batch, nsrc, nfilters, nframes)$ Nrrr() output_size)rFr*rGshaperHunfoldsizer=rKlenrJ intra_process inter_processrPreshaper;r:foldrSrUrV) r mixture_w n_orig_framesbatch n_filtersn_chunks layer_idxintrainteroutputest_maskr&r&r'r.s"      zDPTransformer.forwardc Cs:|j|j|j|j|j|j|j|j|j|j |j |j d }|S)N r:r<r r=r>r?r;r@rArBrrrl)rconfigr&r&r' get_configszDPTransformer.get_config) r5r6r7Nr8r r r Tr)r/r0r1r2rr.rnr3r&r&r$r'r4FsV r4)mathrrDtorch.nnrtorch.nn.modules.activationrrrtorchutilsrdsp.overlap_addr Moduler r4r&r&r&r's     :