o ߥi @sddlZddlZddlmZddlmmZGdddejZ Gddde Z de de de d e fd d Z Gd d d ejZ Gddde ZGdddejjZdS)NcsVeZdZ         dfdd Zd d Zd ed ed efddZddZZS)Conv1d_ONsameTreflectFc st||_||_||_||_| |_d|_| |_|dur&|dur&t d|dur/| |}t j |||j|j|jd|| d|_ dS)NFz.Must provide one of input_shape or in_channelsr)stridedilationpaddinggroupsbias)super__init__ kernel_sizerrr padding_mode unsqueezeskip_transpose ValueError_check_input_shapennConv1dconv) self out_channelsr input_shape in_channelsrrrr r rr __class__S/home/ubuntu/.local/lib/python3.10/site-packages/modelscope/models/audio/sv/TDNN.pyr s,   zConv1d_O.__init__cCs|js |dd}|jr|d}|jdkr"|||j|j|j}n#|jdkr8|jd|j}t ||df}n |jdkr>nt d|j| |}|jrR| d}|js[|dd}|S)zReturns the output of the convolution. Arguments --------- x : torch.Tensor (batch, time, channel) input to convolve. 2d or 4d tensors are expected. rrcausalrvalidz1Padding must be 'same', 'valid' or 'causal'. Got ) r transposerr_manage_paddingr rrFpadrrsqueeze)rxnum_padwxrrrforward3s0        zConv1d_O.forwardr rrcCs.|jd}t||||}tj|||jd}|S)Nr)mode)shapeget_padding_elemr#r$r)rr&r rrL_inrrrrr"\s zConv1d_O._manage_paddingcCsrt|dkr d|_d}n|jr|d}nt|dkr|d}n tdtt||jddkr7td|j|S)zIChecks the input shape and returns the number of input channels. Trz"conv1d expects 2d, 3d inputs. Got rz4The field kernel size must be an odd number. Got %s.)lenrrrstrr )rr+rrrrrns"     zConv1d_O._check_input_shape) NNrrrrTrF) __name__ __module__ __qualname__r r)intr"r __classcell__rrrrr s(() rceZdZfddZZS)rctj|ddi|dSNrTr r rargskwargsrrrr zConv1d.__init__r2r3r4r r6rrrrrrr-rr rcCs|dkr%t||||d}||d||}|d|dg}|S|||dd|d}||d||dg}|S)zThis function computes the number of elements to add for zero-padding. Arguments --------- L_in : int stride: int kernel_size : int dilation : int rr.)mathceil)r-rr rn_stepsL_outrrrrr,s r,cs6eZdZ        d fdd Zdd ZZS) BatchNorm1d_ONh㈵>皙?TFc sVt||_||_|dur|r|d}n|dur|d}tj|||||d|_dS)Nrr)epsmomentumaffinetrack_running_stats)r r combine_batch_timerr BatchNorm1dnorm) rr input_sizerHrIrJrKrLrrrrr s    zBatchNorm1d_O.__init__cCs|j}|jr,|jdkr||d|d|d}n||d|d|d|d}n |js5|dd}||}|jrD||}|S|jsM|dd}|S)aReturns the normalized input tensor. Arguments --------- x : torch.Tensor (batch, time, [channels]) input to normalize. 2d or 3d tensors are expected in input 4d tensors can be used when combine_dims=True. r/rrr.r)r+rLndimreshaperr!rN)rr&shape_orx_nrrrr)s       zBatchNorm1d_O.forward)NNrFrGTTFF)r2r3r4r r)r6rrrrrEsrEcr7)rMcr8r9r:r;rrrr r>zBatchNorm1d.__init__r?rrrrrMr@rMcsLeZdZdZdejjdgdgdgdddffd d Zdd d ZZ S)XvectoraLThis model extracts X-vectors for speaker recognition and diarization. Arguments --------- device : str Device used e.g. "cpu" or "cuda". activation : torch class A class for constructing the activation layers. tdnn_blocks : int Number of time-delay neural (TDNN) layers. tdnn_channels : list of ints Output channels for TDNN layer. tdnn_kernel_sizes : list of ints List of kernel sizes for each TDNN layer. tdnn_dilations : list of ints List of dilations for kernels in each TDNN layer. lin_neurons : int Number of neurons in linear layers. Example ------- >>> compute_xvect = Xvector('cpu') >>> input_feats = torch.rand([5, 10, 40]) >>> outputs = compute_xvect(input_feats) >>> outputs.shape torch.Size([5, 1, 512]) cpu)rWrWrWi)rVr/r/rr)rr.r/rrrWPc sdtt|_t|D]!} || } |jt|| || || d|t| dg|| }qdS)N)rrr r)rO) r r r ModuleListblocksrangeextendrrM) rdevice activation tdnn_blocks tdnn_channelstdnn_kernel_sizestdnn_dilations lin_neuronsr block_indexrrrrr s   zXvector.__init__Nc CsR|dd}|jD]}z|||d}Wq ty ||}Yq w|dd}|S)z]Returns the x-vectors. Arguments --------- x : torch.Tensor rr.)lengths)r!rZ TypeError)rr&lenslayerrrrr)s     zXvector.forward)N) r2r3r4__doc__torchr LeakyReLUr r)r6rrrrrTsrT)numpynprjtorch.nnrtorch.nn.functional functionalr#Modulerrr5r,rErMrTrrrrs }<