o i @s*ddlZddlmZGdddejZdS)N)nnc sJeZdZdZd dedededeffdd Zd ejd ejfd d Z Z S)ISTFTa Custom implementation of ISTFT since torch.istft doesn't allow custom padding (other than `center=True`) with windowing. This is because the NOLA (Nonzero Overlap Add) check fails at the edges. See issue: https://github.com/pytorch/pytorch/issues/62323 Specifically, in the context of neural vocoding we are interested in "same" padding analogous to CNNs. The NOLA constraint is met as we trim padded samples anyway. Args: n_fft (int): Size of Fourier transform. hop_length (int): The distance between neighboring sliding window frames. win_length (int): The size of window frame and STFT filter. padding (str, optional): Type of padding. Options are "center" or "same". Defaults to "same". samen_fft hop_length win_lengthpaddingcsLt|dvr td||_||_||_||_t|}| d|dS)N)centerr#Padding must be 'center' or 'same'.window) super__init__ ValueErrorrrrrtorch hann_windowregister_buffer)selfrrrrr  __class__N/home/ubuntu/.local/lib/python3.10/site-packages/soprano/vocos/spectral_ops.pyr s  zISTFT.__init__specreturnc Csv|jdkr$d|dddf<d|dddf<tj||j|j|j|jddS|jdkr2|j|jd}ntd |d ks@Jd |j \}}}tj j ||jd d d}||jddddf}|d |j|j}tj j j|d |fd |jfd |jfddddd|| f}|jd |dd d} tj j j| d |fd |jfd |jfd|| } | dksJ|| }|S)a Compute the Inverse Short Time Fourier Transform (ISTFT) of a complex spectrogram. Args: spec (Tensor): Input complex spectrogram of shape (B, N, T), where B is the batch size, N is the number of frequency bins, and T is the number of time frames. Returns: Tensor: Reconstructed time-domain signal of shape (B, L), where L is the length of the output signal. r rNT)r rr zExpected a 3D tensor as inputbackward)dimnorm) output_size kernel_sizestridegdy=)rristftrrrr rrshapefftirfftr functionalfoldsquareexpand transposesqueezeall) rrpadBNTifftr y window_sqwindow_enveloperrrforwards4    z ISTFT.forward)r) __name__ __module__ __qualname____doc__intstrr rTensorr6 __classcell__rrrrrs  r)rrModulerrrrrs