espnet2.layers.time_warp.time_warp

source

espnet2.layers.time_warp.time_warp(x: Tensor, window: int = 80, mode: str = 'bicubic')

Time warp module.

This module provides functions and classes for performing time warping on tensor data using PyTorch’s interpolation capabilities.

The time_warp function applies a time warping effect to the input tensor x based on the specified window size and interpolation mode. The TimeWarp class encapsulates this functionality, allowing for easy integration into neural network models.

espnet2.layers.time_warp.DEFAULT_TIME_WARP_MODE

The default interpolation mode, set to

• Type: str

"bicubic".

• Parameters:
  • x (torch.Tensor) – Input tensor of shape (Batch, Time, Freq).
  • window (int) – Time warp parameter that defines the range of the warp.
  • 80. (Default is)
  • mode (str) – Interpolation mode to be used for warping. Default is
  • DEFAULT_TIME_WARP_MODE.
• Returns: The time-warped tensor of the same shape as the input.
• Return type: torch.Tensor
• Raises:ValueError – If the input tensor does not have at least 3 dimensions.

Examples

>>> import torch
>>> x = torch.rand(2, 100, 64)  # (Batch, Time, Freq)
>>> warped_x = time_warp(x, window=50, mode='linear')

>>> time_warp_layer = TimeWarp(window=50, mode='linear')
>>> output, lengths = time_warp_layer(x)

NOTE

Bicubic interpolation supports tensors with 4 or more dimensions.