o si6@sddlmZddlmZddlmZddlmZmZddl m Z m Z ddl m Z dd lZGd d d eZ  dddZd S))nn)BaseEncoderMaskerDecoder) make_enc_dec)magmagreim)norms activations) pad_x_to_yNcsbeZdZdZ           dfd d Zd dZddZddZddZddZ Z S)DeMaska Simple MLP model for surgical mask speech enhancement A transformed-domain masking approach is used. Args: input_type (str, optional): whether the magnitude spectrogram "mag" or both real imaginary parts "reim" are passed as features to the masker network. Concatenation of "mag" and "reim" also can be used by using "cat". output_type (str, optional): whether the masker ouputs a mask for magnitude spectrogram "mag" or both real imaginary parts "reim". hidden_dims (list, optional): list of MLP hidden layer sizes. dropout (float, optional): dropout probability. activation (str, optional): type of activation used in hidden MLP layers. mask_act (str, optional): Which non-linear function to generate mask. norm_type (str, optional): To choose from ``'BN'``, ``'gLN'``, ``'cLN'``. fb_name (str): type of analysis and synthesis filterbanks used, choose between ["stft", "free", "analytic_free"]. n_filters (int): number of filters in the analysis and synthesis filterbanks. stride (int): filterbank filters stride. kernel_size (int): length of filters in the filterbank. encoder_activation (str) sample_rate (float): Sampling rate of the model. **fb_kwargs (dict): Additional kwards to pass to the filterbank creation. rirelugLNstft>c  st|f| | | | d| \}}|||j}|||j}t|||||||d}t|||||_||_||_ ||_ ||_ ||_ ||_ dS)N) kernel_size n_filtersstride sample_rate) norm_type activation hidden_dimsdropoutmask_act)r_get_n_feats_input n_feats_out_get_n_feats_outputbuild_demask_maskersuper__init__ input_type output_typerrrrr)selfr$r%rrrrrfb_namerrrr fb_kwargsencoderdecoder n_masker_in n_masker_outmasker __class__J/home/ubuntu/.local/lib/python3.10/site-packages/asteroid/models/demask.pyr#'s:    zDeMask.__init__cCs8|dkr|S|dvrtd|d}|dkr||7}|S)Nreim>catrz,Input type should be either mag, reim or catrr3NotImplementedError)r&r$ encoder_n_out n_feats_inputr0r0r1rUszDeMask._get_n_feats_inputcCs$|dkr|dS|dkr|Std)Nrrr2z(Output type should be either mag or reimr4)r&r%r6r0r0r1r as zDeMask._get_n_feats_outputcCsP|}|jdkr t|}n |jdkrt|}||}|jdkr&|ddd}|S)a Estimates masks based on time-frequency representations. Args: tf_rep (torch.Tensor): Time-frequency representation in (batch, freq, seq). Returns: torch.Tensor: Estimated masks in (batch, freq, seq). rr3rr)r$rrr-r%repeat)r&tf_rep masker_input est_masksr0r0r1forward_maskerhs     zDeMask.forward_maskercCs|jdkr |d}||S)aQApplies masks to time-frequency representations. Args: tf_rep (torch.Tensor): Time-frequency representations in (batch, freq, seq). est_masks (torch.Tensor): Estimated masks in (batch, freq, seq). Returns: torch.Tensor: Masked time-frequency representations. r2r)r% unsqueeze)r&r9r;r0r0r1 apply_masks|s zDeMask.apply_maskscCs8|j|j|j|j|j|j|jd}||jj |S)z-Arguments needed to re-instantiate the model.)r$r%rrrrr) r$r%rrrrrupdater) filterbank get_config)r& model_argsr0r0r1get_model_argss zDeMask.get_model_args) rrr rrrrrrrrr) __name__ __module__ __qualname____doc__r#rr r<r>rC __classcell__r0r0r.r1r s(. r rrr rc Cst|}||g}t|} |} |D]} |t| | d|| | t|g| } q|t| |dt|gtj|S)Nr)r getr extendrConv1dDropout Sequential) n_inn_outrrrrrmake_layer_normnetlayer_activationin_chan hidden_dimr0r0r1r!s     r!)rrr rr)torchr base_modelsrasteroid_filterbanksrasteroid_filterbanks.transformsrrmasknnr r utils.torch_utilsr warningsr r!r0r0r0r1s