espnet2.svs.feats_extract.score_feats_extract.expand_to_frame

source

espnet2.svs.feats_extract.score_feats_extract.expand_to_frame(expand_len, len_size, label, midi, duration)

Expand the phone-level features to frame-level features.

This function takes the expansion lengths for each phone and replicates the corresponding labels, midi, and duration features to create a frame-level representation. It returns the expanded sequences along with their lengths.

• Parameters:
  • expand_len (List *[*List *[*int ] ]) – A list of lists containing the number of frames each phone should be expanded to for each sample in the batch.
  • len_size (List *[*int ]) – A list containing the sizes of the phone sequences for each sample in the batch.
  • label (torch.Tensor) – A tensor of shape (Batch, Max_Phone_Length) containing the phone labels.
  • midi (torch.Tensor) – A tensor of shape (Batch, Max_Phone_Length) containing the midi values corresponding to the phones.
  • duration (torch.Tensor) – A tensor of shape (Batch, Max_Phone_Length) containing the duration values corresponding to the phones.
• Returns: Tuple[torch.Tensor, torch.Tensor, torch.Tensor, torch.Tensor, torch.Tensor, torch.Tensor]: A tuple containing:

  • Expanded label tensor of shape (Batch, Expanded_Length).
  • Lengths of the expanded labels tensor.
  • Expanded midi tensor of shape (Batch, Expanded_Length).
  • Lengths of the expanded midi tensor.
  • Expanded duration tensor of shape (Batch, Expanded_Length).
  • Lengths of the expanded duration tensor.

Examples

>>> expand_len = [[2, 3], [1, 4]]
>>> len_size = [2, 2]
>>> label = torch.tensor([[1, 2], [3, 4]])
>>> midi = torch.tensor([[60, 62], [64, 65]])
>>> duration = torch.tensor([[100, 200], [300, 400]])
>>> result = expand_to_frame(expand_len, len_size, label, midi, duration)
>>> print(result)
(tensor([[1, 1, 2, 2, 2],
          [3, 4, 4, 4, 4]]),
 tensor([5, 5]),
 tensor([[60, 60, 62, 62, 62],
          [64, 65, 65, 65, 65]]),
 tensor([5, 5]),
 tensor([[100, 100, 200, 200, 200],
          [300, 400, 400, 400, 400]]),
 tensor([5, 5]))