Shortcuts

DNN building blocks

Convolutional blocks

class asteroid.masknn.convolutional.Conv1DBlock(in_chan, hid_chan, skip_out_chan, kernel_size, padding, dilation, norm_type='gLN', causal=False)[source]

Bases: torch.nn.Module

One dimensional convolutional block, as proposed in [1].

Parameters
  • in_chan (int) – Number of input channels.

  • hid_chan (int) – Number of hidden channels in the depth-wise convolution.

  • skip_out_chan (int) – Number of channels in the skip convolution. If 0 or None, Conv1DBlock won’t have any skip connections. Corresponds to the the block in v1 or the paper. The forward return res instead of [res, skip] in this case.

  • kernel_size (int) – Size of the depth-wise convolutional kernel.

  • padding (int) – Padding of the depth-wise convolution.

  • dilation (int) – Dilation of the depth-wise convolution.

  • norm_type (str, optional) –

    Type of normalization to use. To choose from

    • 'gLN': global Layernorm.

    • 'cLN': channelwise Layernorm.

    • 'cgLN': cumulative global Layernorm.

    • Any norm supported by get()

  • causal (bool, optional) – Whether or not the convolutions are causal

References

[1] : “Conv-TasNet: Surpassing ideal time-frequency magnitude masking for speech separation” TASLP 2019 Yi Luo, Nima Mesgarani https://arxiv.org/abs/1809.07454

forward(x)[source]

Input shape $(batch, feats, seq)$.

class asteroid.masknn.convolutional.TDConvNet(in_chan, n_src, out_chan=None, n_blocks=8, n_repeats=3, bn_chan=128, hid_chan=512, skip_chan=128, conv_kernel_size=3, norm_type='gLN', mask_act='relu', causal=False)[source]

Bases: torch.nn.Module

Temporal Convolutional network used in ConvTasnet.

Parameters
  • in_chan (int) – Number of input filters.

  • n_src (int) – Number of masks to estimate.

  • out_chan (int, optional) – Number of bins in the estimated masks. If None, out_chan = in_chan.

  • n_blocks (int, optional) – Number of convolutional blocks in each repeat. Defaults to 8.

  • n_repeats (int, optional) – Number of repeats. Defaults to 3.

  • bn_chan (int, optional) – Number of channels after the bottleneck.

  • hid_chan (int, optional) – Number of channels in the convolutional blocks.

  • skip_chan (int, optional) – Number of channels in the skip connections. If 0 or None, TDConvNet won’t have any skip connections and the masks will be computed from the residual output. Corresponds to the ConvTasnet architecture in v1 or the paper.

  • conv_kernel_size (int, optional) – Kernel size in convolutional blocks.

  • norm_type (str, optional) – To choose from 'BN', 'gLN', 'cLN'.

  • mask_act (str, optional) – Which non-linear function to generate mask.

  • causal (bool, optional) – Whether or not the convolutions are causal.

References

[1] : “Conv-TasNet: Surpassing ideal time-frequency magnitude masking for speech separation” TASLP 2019 Yi Luo, Nima Mesgarani https://arxiv.org/abs/1809.07454

forward(mixture_w)[source]

Forward.

Parameters

mixture_w (torch.Tensor) – Tensor of shape $(batch, nfilters, nframes)$

Returns

torch.Tensor – estimated mask of shape $(batch, nsrc, nfilters, nframes)$

class asteroid.masknn.convolutional.TDConvNetpp(in_chan, n_src, out_chan=None, n_blocks=8, n_repeats=3, bn_chan=128, hid_chan=512, skip_chan=128, conv_kernel_size=3, norm_type='fgLN', mask_act='relu')[source]

Bases: torch.nn.Module

Improved Temporal Convolutional network used in [1] (TDCN++)

Parameters
  • in_chan (int) – Number of input filters.

  • n_src (int) – Number of masks to estimate.

  • out_chan (int, optional) – Number of bins in the estimated masks. If None, out_chan = in_chan.

  • n_blocks (int, optional) – Number of convolutional blocks in each repeat. Defaults to 8.

  • n_repeats (int, optional) – Number of repeats. Defaults to 3.

  • bn_chan (int, optional) – Number of channels after the bottleneck.

  • hid_chan (int, optional) – Number of channels in the convolutional blocks.

  • skip_chan (int, optional) – Number of channels in the skip connections. If 0 or None, TDConvNet won’t have any skip connections and the masks will be computed from the residual output. Corresponds to the ConvTasnet architecture in v1 or the paper.

  • kernel_size (int, optional) – Kernel size in convolutional blocks.

  • norm_type (str, optional) – To choose from 'BN', 'gLN', 'cLN'.

  • mask_act (str, optional) – Which non-linear function to generate mask.

References

[1] : Kavalerov, Ilya et al. “Universal Sound Separation.” in WASPAA 2019

Note

The differences wrt to ConvTasnet’s TCN are:

  1. Channel wise layer norm instead of global

  2. Longer-range skip-residual connections from earlier repeat inputs to later repeat inputs after passing them through dense layer.

  3. Learnable scaling parameter after each dense layer. The scaling parameter for the second dense layer in each convolutional block (which is applied rightbefore the residual connection) is initialized to an exponentially decaying scalar equal to 0.9**L, where L is the layer or block index.

forward(mixture_w)[source]

Forward.

Parameters

mixture_w (torch.Tensor) – Tensor of shape $(batch, nfilters, nframes)$

Returns

torch.Tensor – estimated mask of shape $(batch, nsrc, nfilters, nframes)$

class asteroid.masknn.convolutional.DCUNetComplexEncoderBlock(in_chan, out_chan, kernel_size, stride, padding, norm_type='bN', activation='leaky_relu')[source]

Bases: torch.nn.Module

Encoder block as proposed in [1].

Parameters
  • in_chan (int) – Number of input channels.

  • out_chan (int) – Number of output channels.

  • kernel_size (Tuple[int, int]) – Convolution kernel size.

  • stride (Tuple[int, int]) – Convolution stride.

  • padding (Tuple[int, int]) – Convolution padding.

  • norm_type (str, optional) – Type of normalization to use. See norms for valid values.

  • activation (str, optional) – Type of activation to use. See activations for valid values.

References

[1] : “Phase-aware Speech Enhancement with Deep Complex U-Net”, Hyeong-Seok Choi et al. https://arxiv.org/abs/1903.03107

class asteroid.masknn.convolutional.DCUNetComplexDecoderBlock(in_chan, out_chan, kernel_size, stride, padding, output_padding=(0, 0), norm_type='bN', activation='leaky_relu')[source]

Bases: torch.nn.Module

Decoder block as proposed in [1].

Parameters
  • in_chan (int) – Number of input channels.

  • out_chan (int) – Number of output channels.

  • kernel_size (Tuple[int, int]) – Convolution kernel size.

  • stride (Tuple[int, int]) – Convolution stride.

  • padding (Tuple[int, int]) – Convolution padding.

  • norm_type (str, optional) – Type of normalization to use. See norms for valid values.

  • activation (str, optional) – Type of activation to use. See activations for valid values.

References

[1] : “Phase-aware Speech Enhancement with Deep Complex U-Net”, Hyeong-Seok Choi et al. https://arxiv.org/abs/1903.03107

class asteroid.masknn.convolutional.DCUMaskNet(encoders, decoders, fix_length_mode=None, **kwargs)[source]

Bases: asteroid.masknn.base.BaseDCUMaskNet

Masking part of DCUNet, as proposed in [1].

Valid architecture values for the default_architecture classmethod are: “Large-DCUNet-20”, “DCUNet-20”, “DCUNet-16”, “DCUNet-10” and “mini”.

Valid fix_length_mode values are [None, “pad”, “trim”].

Input shape is expected to be $(batch, nfreqs, time)$, with $nfreqs - 1$ divisible by $f_0 * f_1 * … * f_N$ where $f_k$ are the frequency strides of the encoders, and $time - 1$ is divisible by $t_0 * t_1 * … * t_N$ where $t_N$ are the time strides of the encoders.

References

[1] : “Phase-aware Speech Enhancement with Deep Complex U-Net”, Hyeong-Seok Choi et al. https://arxiv.org/abs/1903.03107

fix_input_dims(x)[source]

Overwrite this in subclasses to implement input dimension checks.

fix_output_dims(out, x)[source]

Overwrite this in subclasses to implement output dimension checks. y is the output and x was the input (passed to use the shape).

class asteroid.masknn.convolutional.SuDORMRF(in_chan, n_src, bn_chan=128, num_blocks=16, upsampling_depth=4, mask_act='softmax')[source]

Bases: torch.nn.Module

SuDORMRF mask network, as described in [1].

Parameters
  • in_chan (int) – Number of input channels. Also number of output channels.

  • n_src (int) – Number of sources in the input mixtures.

  • bn_chan (int, optional) – Number of bins in the bottleneck layer and the UNet blocks.

  • num_blocks (int) – Number of of UBlocks.

  • upsampling_depth (int) – Depth of upsampling.

  • mask_act (str) – Name of output activation.

References

[1] : “Sudo rm -rf: Efficient Networks for Universal Audio Source Separation”, Tzinis et al. MLSP 2020.

class asteroid.masknn.convolutional.SuDORMRFImproved(in_chan, n_src, bn_chan=128, num_blocks=16, upsampling_depth=4, mask_act='relu')[source]

Bases: torch.nn.Module

Improved SuDORMRF mask network, as described in [1].

Parameters
  • in_chan (int) – Number of input channels. Also number of output channels.

  • n_src (int) – Number of sources in the input mixtures.

  • bn_chan (int, optional) – Number of bins in the bottleneck layer and the UNet blocks.

  • num_blocks (int) – Number of of UBlocks

  • upsampling_depth (int) – Depth of upsampling

  • mask_act (str) – Name of output activation.

References

[1] : “Sudo rm -rf: Efficient Networks for Universal Audio Source Separation”, Tzinis et al. MLSP 2020.

class asteroid.masknn.convolutional.UBlock(out_chan=128, in_chan=512, upsampling_depth=4)[source]

Bases: asteroid.masknn.convolutional._BaseUBlock

Upsampling block.

Based on the following principle: REDUCE ---> SPLIT ---> TRANSFORM --> MERGE

forward(x)[source]
Parameters

x – input feature map

Returns

transformed feature map

class asteroid.masknn.convolutional.UConvBlock(out_chan=128, in_chan=512, upsampling_depth=4)[source]

Bases: asteroid.masknn.convolutional._BaseUBlock

Block which performs successive downsampling and upsampling in order to be able to analyze the input features in multiple resolutions.

forward(x)[source]
Args

x: input feature map

Returns

transformed feature map

Recurrent blocks

class asteroid.masknn.recurrent.SingleRNN(rnn_type, input_size, hidden_size, n_layers=1, dropout=0, bidirectional=False)[source]

Bases: torch.nn.Module

Module for a RNN block.

Inspired from https://github.com/yluo42/TAC/blob/master/utility/models.py Licensed under CC BY-NC-SA 3.0 US.

Parameters
  • 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.

  • dropout (float, optional) – Dropout ratio. Default is 0.

  • bidirectional (bool, optional) – Whether the RNN layers are bidirectional. Default is False.

forward(inp)[source]

Input shape [batch, seq, feats]

class asteroid.masknn.recurrent.MulCatRNN(rnn_type, input_size, hidden_size, n_layers=1, dropout=0, bidirectional=False)[source]

Bases: torch.nn.Module

MulCat RNN block from [1].

Composed of two RNNs, returns cat([RNN_1(x) * RNN_2(x), x]).

Parameters
  • 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.

  • dropout (float, optional) – Dropout ratio. Default is 0.

  • bidirectional (bool, optional) – Whether the RNN layers are bidirectional. Default is False.

References

[1] Eliya Nachmani, Yossi Adi, & Lior Wolf. (2020). Voice Separation with an Unknown Number of Multiple Speakers.

forward(inp)[source]

Input shape [batch, seq, feats]

class asteroid.masknn.recurrent.StackedResidualRNN(rnn_type, n_units, n_layers=4, dropout=0.0, bidirectional=False)[source]

Bases: torch.nn.Module

Stacked RNN with builtin residual connection. Only supports forward RNNs. See StackedResidualBiRNN for bidirectional ones.

Parameters
  • rnn_type (str) – Select from 'RNN', 'LSTM', 'GRU'. Can also be passed in lowercase letters.

  • n_units (int) – Number of units in recurrent layers. This will also be the expected input size.

  • n_layers (int) – Number of recurrent layers.

  • dropout (float) – Dropout value, between 0. and 1. (Default: 0.)

  • bidirectional (bool) – If True, use bidirectional RNN, else unidirectional. (Default: False)

forward(x)[source]

Builtin residual connections + dropout applied before residual. Input shape : [batch, time_axis, feat_axis]

class asteroid.masknn.recurrent.StackedResidualBiRNN(rnn_type, n_units, n_layers=4, dropout=0.0, bidirectional=True)[source]

Bases: torch.nn.Module

Stacked Bidirectional RNN with builtin residual connection. Residual connections are applied on both RNN directions. Only supports bidiriectional RNNs. See StackedResidualRNN for unidirectional ones.

Parameters
  • rnn_type (str) – Select from 'RNN', 'LSTM', 'GRU'. Can also be passed in lowercase letters.

  • n_units (int) – Number of units in recurrent layers. This will also be the expected input size.

  • n_layers (int) – Number of recurrent layers.

  • dropout (float) – Dropout value, between 0. and 1. (Default: 0.)

  • bidirectional (bool) – If True, use bidirectional RNN, else unidirectional. (Default: False)

forward(x)[source]

Builtin residual connections + dropout applied before residual. Input shape : [batch, time_axis, feat_axis]

class asteroid.masknn.recurrent.DPRNNBlock(in_chan, hid_size, norm_type='gLN', bidirectional=True, rnn_type='LSTM', use_mulcat=False, num_layers=1, dropout=0)[source]

Bases: torch.nn.Module

Dual-Path RNN Block as proposed in [1].

Parameters
  • in_chan (int) – Number of input channels.

  • hid_size (int) – Number of hidden neurons in the RNNs.

  • norm_type (str, optional) – Type of normalization to use. To choose from - 'gLN': global Layernorm - 'cLN': channelwise Layernorm

  • bidirectional (bool, optional) – True for bidirectional Inter-Chunk RNN.

  • rnn_type (str, optional) – Type of RNN used. Choose from 'RNN', 'LSTM' and 'GRU'.

  • num_layers (int, optional) – Number of layers used in each RNN.

  • dropout (float, optional) – Dropout ratio. Must be in [0, 1].

References

[1] “Dual-path RNN: efficient long sequence modeling for time-domain single-channel speech separation”, Yi Luo, Zhuo Chen and Takuya Yoshioka. https://arxiv.org/abs/1910.06379

forward(x)[source]

Input shape : [batch, feats, chunk_size, num_chunks]

class asteroid.masknn.recurrent.DPRNN(in_chan, n_src, out_chan=None, bn_chan=128, hid_size=128, chunk_size=100, hop_size=None, n_repeats=6, norm_type='gLN', mask_act='relu', bidirectional=True, rnn_type='LSTM', use_mulcat=False, num_layers=1, dropout=0)[source]

Bases: torch.nn.Module

Dual-path RNN Network for Single-Channel Source Separation

introduced in [1].

Parameters
  • in_chan (int) – Number of input filters.

  • n_src (int) – Number of masks to estimate.

  • out_chan (int or None) – Number of bins in the estimated masks. Defaults to in_chan.

  • bn_chan (int) – Number of channels after the bottleneck. Defaults to 128.

  • hid_size (int) – Number of neurons in the RNNs cell state. Defaults to 128.

  • 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. To choose from

    • 'gLN': global Layernorm

    • 'cLN': channelwise Layernorm

  • 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).

  • rnn_type (str, optional) – Type of RNN used. Choose between 'RNN', 'LSTM' and 'GRU'.

  • num_layers (int, optional) – Number of layers in each RNN.

  • dropout (float, optional) – Dropout ratio, must be in [0,1].

References

[1] “Dual-path RNN: efficient long sequence modeling for time-domain single-channel speech separation”, Yi Luo, Zhuo Chen and Takuya Yoshioka. https://arxiv.org/abs/1910.06379

forward(mixture_w)[source]

Forward.

Parameters

mixture_w (torch.Tensor) – Tensor of shape $(batch, nfilters, nframes)$

Returns

torch.Tensor – estimated mask of shape $(batch, nsrc, nfilters, nframes)$

class asteroid.masknn.recurrent.LSTMMasker(in_chan, n_src, out_chan=None, rnn_type='lstm', n_layers=4, hid_size=512, dropout=0.3, mask_act='sigmoid', bidirectional=True)[source]

Bases: torch.nn.Module

LSTM mask network introduced in [1], without skip connections.

Parameters
  • in_chan (int) – Number of input filters.

  • n_src (int) – Number of masks to estimate.

  • out_chan (int or None) – Number of bins in the estimated masks. Defaults to in_chan.

  • rnn_type (str, optional) – Type of RNN used. Choose between 'RNN', 'LSTM' and 'GRU'.

  • n_layers (int, optional) – Number of layers in each RNN.

  • hid_size (int) – Number of neurons in the RNNs cell state.

  • mask_act (str, optional) – Which non-linear function to generate mask.

  • bidirectional (bool, optional) – Whether to use BiLSTM

  • dropout (float, optional) – Dropout ratio, must be in [0,1].

References

[1]: Yi Luo et al. “Real-time Single-channel Dereverberation and Separation with Time-domain Audio Separation Network”, Interspeech 2018

class asteroid.masknn.recurrent.DCCRMaskNetRNN(in_size, hid_size=128, rnn_type='LSTM', n_layers=2, norm_type=None, **rnn_kwargs)[source]

Bases: torch.nn.Module

RNN (LSTM) layer between encoders and decoders introduced in [1].

Parameters
  • in_size (int) – Number of inputs to the RNN. Must be the product of non-batch, non-time dimensions of output shape of last encoder, i.e. if the last encoder output shape is $(batch, nchans, nfreqs, time)$, in_size must be $nchans * nfreqs$.

  • hid_size (int, optional) – Number of units in RNN.

  • rnn_type (str, optional) – Type of RNN to use. See SingleRNN for valid values.

  • n_layers (int, optional) – Number of layers used in RNN.

  • norm_type (Optional[str], optional) – Norm to use after linear. See asteroid.masknn.norms for valid values. (Not used in [1]).

  • rnn_kwargs (optional) – Passed to SingleRNN().

References

[1] : “DCCRN: Deep Complex Convolution Recurrent Network for Phase-Aware Speech Enhancement”, Yanxin Hu et al. https://arxiv.org/abs/2008.00264

forward(x: torch.Tensor)[source]

Input shape: [batch, …, time]

class asteroid.masknn.recurrent.DCCRMaskNet(encoders, decoders, n_freqs, **kwargs)[source]

Bases: asteroid.masknn.base.BaseDCUMaskNet

Masking part of DCCRNet, as proposed in [1].

Valid architecture values for the default_architecture classmethod are: “DCCRN” and “mini”.

Parameters
  • encoders (list of length N of tuples of (in_chan, out_chan, kernel_size, stride, padding)) – Arguments of encoders of the u-net

  • decoders (list of length N of tuples of (in_chan, out_chan, kernel_size, stride, padding)) – Arguments of decoders of the u-net

  • n_freqs (int) – Number of frequencies (dim 1) of input to .forward(). Must be divisible by $f_0 * f_1 * … * f_N$ where $f_k$ are the frequency strides of the encoders.

Input shape is expected to be $(batch, nfreqs, time)$, with $nfreqs$ divisible by $f_0 * f_1 * … * f_N$ where $f_k$ are the frequency strides of the encoders.

References

[1] : “DCCRN: Deep Complex Convolution Recurrent Network for Phase-Aware Speech Enhancement”, Yanxin Hu et al. https://arxiv.org/abs/2008.00264

fix_input_dims(x)[source]

Overwrite this in subclasses to implement input dimension checks.

Attention blocks

class asteroid.masknn.attention.ImprovedTransformedLayer(embed_dim, n_heads, dim_ff, dropout=0.0, activation='relu', bidirectional=True, norm='gLN')[source]

Bases: torch.nn.Module

Improved Transformer module as used in [1]. It is Multi-Head self-attention followed by LSTM, activation and linear projection layer.

Parameters
  • 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).

class asteroid.masknn.attention.DPTransformer(in_chan, n_src, n_heads=4, ff_hid=256, chunk_size=100, hop_size=None, n_repeats=6, norm_type='gLN', ff_activation='relu', mask_act='relu', bidirectional=True, dropout=0)[source]

Bases: torch.nn.Module

Dual-path Transformer introduced in [1].

Parameters
  • 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).

forward(mixture_w)[source]

Forward.

Parameters

mixture_w (torch.Tensor) – Tensor of shape $(batch, nfilters, nframes)$

Returns

torch.Tensor – estimated mask of shape $(batch, nsrc, nfilters, nframes)$

Norms

class asteroid.masknn.norms.GlobLN(channel_size)[source]

Bases: asteroid.masknn.norms._LayerNorm

Global Layer Normalization (globLN).

forward(x, EPS: float = 1e-08)[source]

Applies forward pass.

Works for any input size > 2D.

Parameters

x (torch.Tensor) – Shape [batch, chan, *]

Returns

torch.Tensor – gLN_x [batch, chan, *]

class asteroid.masknn.norms.ChanLN(channel_size)[source]

Bases: asteroid.masknn.norms._LayerNorm

Channel-wise Layer Normalization (chanLN).

forward(x, EPS: float = 1e-08)[source]

Applies forward pass.

Works for any input size > 2D.

Parameters

x (torch.Tensor) – [batch, chan, *]

Returns

torch.Tensor – chanLN_x [batch, chan, *]

class asteroid.masknn.norms.CumLN(channel_size)[source]

Bases: asteroid.masknn.norms._LayerNorm

Cumulative Global layer normalization(cumLN).

forward(x, EPS: float = 1e-08)[source]
Parameters

x (torch.Tensor) – Shape [batch, channels, length]

Returns

torch.Tensor – cumLN_x [batch, channels, length]

class asteroid.masknn.norms.FeatsGlobLN(channel_size)[source]

Bases: asteroid.masknn.norms._LayerNorm

Feature-wise global Layer Normalization (FeatsGlobLN). Applies normalization over frames for each channel.

forward(x, EPS: float = 1e-08)[source]

Applies forward pass.

Works for any input size > 2D.

Parameters

x (torch.Tensor) – [batch, chan, time]

Returns

torch.Tensor – chanLN_x [batch, chan, time]

class asteroid.masknn.norms.BatchNorm(*args: Any, **kwargs: Any)[source]

Bases: torch.nn.modules.batchnorm._BatchNorm

Wrapper class for pytorch BatchNorm1D and BatchNorm2D

asteroid.masknn.norms.gLN[source]

alias of asteroid.masknn.norms.GlobLN

asteroid.masknn.norms.fgLN[source]

alias of asteroid.masknn.norms.FeatsGlobLN

asteroid.masknn.norms.cLN[source]

alias of asteroid.masknn.norms.ChanLN

asteroid.masknn.norms.cgLN[source]

alias of asteroid.masknn.norms.CumLN

asteroid.masknn.norms.bN[source]

alias of asteroid.masknn.norms.BatchNorm

asteroid.masknn.norms.register_norm(custom_norm)[source]

Register a custom norm, gettable with norms.get.

Parameters

custom_norm – Custom norm to register.

asteroid.masknn.norms.get(identifier)[source]

Returns a norm class from a string. Returns its input if it is callable (already a _LayerNorm for example).

Parameters

identifier (str or Callable or None) – the norm identifier.

Returns

_LayerNorm or None

asteroid.masknn.norms.get_complex(identifier)[source]

Like .get but returns a complex norm created with asteroid.complex_nn.OnReIm.

Complex number support

Complex 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.

asteroid.complex_nn.on_reim(f)[source]

Make a complex-valued function callable from a real-valued one by applying it to the real and imaginary components independently.

Returns

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.

class asteroid.complex_nn.OnReIm(module_cls, *args, **kwargs)[source]

Bases: torch.nn.Module

Like on_reim, but for stateful modules.

Parameters

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.

class asteroid.complex_nn.ComplexMultiplicationWrapper(module_cls, *args, **kwargs)[source]

Bases: torch.nn.Module

Make 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.

Parameters

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.

class asteroid.complex_nn.ComplexSingleRNN(rnn_type, input_size, hidden_size, n_layers=1, dropout=0, bidirectional=False)[source]

Bases: torch.nn.Module

Module 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.

Parameters
  • 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

forward(x: torch.Tensor) → torch.Tensor[source]

Input shape [batch, seq, feats]

class asteroid.complex_nn.BoundComplexMask(bound_type)[source]

Bases: torch.nn.Module

Module version of bound_complex_mask

asteroid.complex_nn.bound_complex_mask(mask: torch.Tensor, bound_type='tanh')[source]

Bound a complex mask, as proposed in [1], section 3.2.

Valid bound types, for a complex mask \(M = |M| ⋅ e^{i φ(M)}\):

  • Unbounded (“UBD”): \(M_{\mathrm{UBD}} = M\)

  • Sigmoid (“BDSS”): \(M_{\mathrm{BDSS}} = σ(|M|) e^{i σ(φ(M))}\)

  • Tanh (“BDT”): \(M_{\mathrm{BDT}} = \mathrm{tanh}(|M|) e^{i φ(M)}\)

Parameters

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