o siE@sdZddlZddlZddlmZddlmZejZddZddZ d d Z d d Z Gd ddej Z Gdddej ZGdddej ZeeejZeeejZGdddej ZddefddZdS)aComplex building blocks that work with PyTorch native (!) complex tensors, i.e. dtypes complex64/complex128, or tensors for which `.is_complex()` returns True. Note that Asteroid code has two other representations of complex numbers: - Torchaudio representation [..., 2] where [..., 0] and [..., 1] are real and imaginary components, respectively - Asteroid style representation, identical to the Torchaudio representation, but with the last dimension concatenated: tensor([r1, r2, ..., rn, i1, i2, ..., in]). The concatenated (2 * n) dimension may be at an arbitrary position, i.e. the tensor is of shape [..., 2 * n, ...]. See `asteroid_filterbanks.transforms` for details. N) transforms)nncCs|SN) is_complexxrG/home/ubuntu/.local/lib/python3.10/site-packages/asteroid/complex_nn.pyis_torch_complexsr cCs,ttj|t||t|fddSN)dim)torchview_as_complexstackcossin)magphaserrr torch_complex_from_magphases$rcCsttj||gddSr )rrr)reimrrr torch_complex_from_reim"srcs>tfdd}|jjdk|jjdk|S)a@Make a complex-valued function callable from a real-valued one by applying it to the real and imaginary components independently. Return: cf(x), complex version of `f`: A function that applies `f` to the real and imaginary components of `x` and returns the result as PyTorch complex tensor. cst|j|jSr)rrealimagrfrr cf/szon_reim..cfz (complex)) functoolswraps__name__ __qualname__)rrrrr on_reim&s  r"cs(eZdZdZfddZddZZS)OnReImzLike `on_reim`, but for stateful modules. Args: module_cls (callable): A class or function that returns a Torch module/functional. Called 2x with *args, **kwargs, to construct the real and imaginary component modules. c.t||i||_||i||_dSrsuper__init__ re_module im_moduleself module_clsargskwargs __class__rr r'C zOnReIm.__init__cCst||j||jSrrr(rr)rr+rrrr forwardHszOnReIm.forward)r __module__r!__doc__r'r4 __classcell__rrr/r r#;s r#cs2eZdZdZfddZdedefddZZS)ComplexMultiplicationWrapperaMake a complex-valued module `F` from a real-valued module `f` by applying complex multiplication rules: F(a + i b) = f1(a) - f1(b) + i (f2(b) + f2(a)) where `f1`, `f2` are instances of `f` that do *not* share weights. Args: module_cls (callable): A class or function that returns a Torch module/functional. Constructor of `f` in the formula above. Called 2x with `*args`, `**kwargs`, to construct the real and imaginary component modules. cr$rr%r*r/rr r'Zr1z%ComplexMultiplicationWrapper.__init__rreturncCs2t||j||j||j||jSrr2r3rrr r4_sz$ComplexMultiplicationWrapper.forwardr r5r!r6r' ComplexTensorr4r7rrr/r r8Ls r8csBeZdZdZ d fdd ZeddZd ed efd d ZZ S)ComplexSingleRNNaModule for a complex RNN block. This is similar to :cls:`asteroid.masknn.recurrent.SingleRNN` but uses complex multiplication as described in [1]. Arguments are identical to those of `SingleRNN`, except for `dropout`, which is not yet supported. Args: rnn_type (str): Select from ``'RNN'``, ``'LSTM'``, ``'GRU'``. Can also be passed in lowercase letters. input_size (int): Dimension of the input feature. The input should have shape [batch, seq_len, input_size]. hidden_size (int): Dimension of the hidden state. n_layers (int, optional): Number of layers used in RNN. Default is 1. bidirectional (bool, optional): Whether the RNN layers are bidirectional. Default is ``False``. dropout: Not yet supported. References [1] : "DCCRN: Deep Complex Convolution Recurrent Network for Phase-Aware Speech Enhancement", Yanxin Hu et al. https://arxiv.org/abs/2008.00264 rFc s|r |dkr Jdtddlm}||dd|d}t|fd|i|} tj| g|_t |dD]} |j t|fd| j j i|q5dS)Nr=z,Dropout is not yet supported for complex RNN) SingleRNNr)rnn_type hidden_sizen_layersdropout bidirectional input_size) r&r'masknn.recurrentr>r8rr ModuleListrnnsrangeappendr( output_size) r+r?rDr@rArBrCr>r. first_rnn_r/rr r'}s,  zComplexSingleRNN.__init__cCs|jdjjS)Nr )rGr(rJ)r+rrr rJszComplexSingleRNN.output_sizerr9cCs|jD]}||}q|S)zInput shape [batch, seq, feats])rG)r+rrnnrrr r4s  zComplexSingleRNN.forward)r=rF) r r5r!r6r'propertyrJr;r4r7rrr/r r<fs r<cs.eZdZdZfddZdefddZZS)BoundComplexMaskz&Module version of `bound_complex_mask`cst||_dSr)r&r' bound_type)r+rPr/rr r's  zBoundComplexMask.__init__maskcCs t||jSr)bound_complex_maskrP)r+rQrrr r4s zBoundComplexMask.forwardr:rrr/r rOs rOtanhrQcCsb|dvr ttj|S|dvr*tt|\}}|dvr#t|}n|}t||Std|)uBound a complex mask, as proposed in [1], section 3.2. Valid bound types, for a complex mask :math:`M = |M| ⋅ e^{i φ(M)}`: - Unbounded ("UBD"): :math:`M_{\mathrm{UBD}} = M` - Sigmoid ("BDSS"): :math:`M_{\mathrm{BDSS}} = σ(|M|) e^{i σ(φ(M))}` - Tanh ("BDT"): :math:`M_{\mathrm{BDT}} = \mathrm{tanh}(|M|) e^{i φ(M)}` Args: bound_type (str or None): The type of bound to use, either of "tanh"/"bdt" (default), "sigmoid"/"bdss" or None/"bdt". References [1] : "Phase-aware Speech Enhancement with Deep Complex U-Net", Hyeong-Seok Choi et al. https://arxiv.org/abs/1903.03107 >BDSSsigmoid>NBDTUBDrS>rVrSzUnknown mask bound ) r"rrUrmagphasefrom_torch_complexrSr ValueError)rQrPmask_mag mask_phasemask_mag_boundedrrr rRs  rR)rS)r6rrasteroid_filterbanksrrTensorr;r rrr"Moduler#r8r<partialConv2d ComplexConv2dConvTranspose2dComplexConvTranspose2drOrRrrrr s"   9