deepextractor.utils.stft
========================

.. py:module:: deepextractor.utils.stft

.. autoapi-nested-parse::

   STFT / iSTFT helpers for magnitude+phase representation.





Module Contents
---------------

.. py:function:: apply_stft(array, n_fft, hop_length, win_length, window)

   Compute STFT and return a magnitude+phase tensor.

   :param array: Input time-series. Batched (N, time) is recommended; a 1D input will
                 produce a (F, 2, time) tensor which is not compatible with UNET2D.
   :type array: array-like, shape (N, time) or (time,)
   :param n_fft: STFT parameters.
   :type n_fft: int
   :param hop_length: STFT parameters.
   :type hop_length: int
   :param win_length: STFT parameters.
   :type win_length: int
   :param window: Analysis window (e.g. Hann), length ``win_length``.
   :type window: torch.Tensor

   :returns: Channel 0 = magnitude, channel 1 = phase.
   :rtype: torch.Tensor, shape (N, 2, F, frames)


.. py:function:: apply_istft(stft_array, n_fft, hop_length, win_length, window)

   Reconstruct time-domain signal from a magnitude+phase tensor.

   :param stft_array: Channel 0 = magnitude, channel 1 = phase.
   :type stft_array: torch.Tensor, shape (N, 2, F, frames)
   :param n_fft: STFT parameters (must match the forward transform).
   :type n_fft: int
   :param hop_length: STFT parameters (must match the forward transform).
   :type hop_length: int
   :param win_length: STFT parameters (must match the forward transform).
   :type win_length: int
   :param window: Synthesis window, moved to the correct device automatically.
   :type window: torch.Tensor

   :rtype: torch.Tensor, shape (N, time)