espnet2.gan_tts.utils.get_random_segments.get_segments

source

espnet2.gan_tts.utils.get_random_segments.get_segments(x: Tensor, start_idxs: Tensor, segment_size: int) → Tensor

Function to get random segments.

This function retrieves random segments from a given tensor based on specified segment sizes and their respective lengths. The output consists of both the segmented tensor and the starting indices of these segments.

• Parameters:
  • x (torch.Tensor) – Input tensor of shape (B, C, T) where:
    • B: Batch size
    • C: Number of channels
    • T: Length of the sequence
  • x_lengths (torch.Tensor) – Length tensor of shape (B,) indicating the valid lengths of each input tensor in the batch.
  • segment_size (int) – The size of the segments to be extracted from the input tensor.
• Returns: A tuple containing: : - Tensor: Segmented tensor of shape (B, C, segment_size).
  • Tensor: Start index tensor of shape (B,) representing the starting indices of each segment.
• Return type: Tuple[torch.Tensor, torch.Tensor]

Examples

>>> x = torch.randn(4, 2, 10)  # A batch of 4 samples, 2 channels, 10 length
>>> x_lengths = torch.tensor([10, 8, 10, 5])  # Lengths of each sample
>>> segment_size = 5
>>> segments, start_idxs = get_random_segments(x, x_lengths, segment_size)
>>> segments.shape
torch.Size([4, 2, 5])  # Segmented tensor shape
>>> start_idxs.shape
torch.Size([4])  # Start index tensor shape