Add some transforms #122
Add some transforms #122
Conversation
| return self.__class__.__name__ + '()' | ||
|
|
||
|
|
||
| class Pad(object): |
There was a problem hiding this comment.
This class seems very similar to PadTrim (
audio/torchaudio/transforms.py
Line 75 in 7ee1d38
Is there any difference?
There was a problem hiding this comment.
does not look like. They bot end up using the same torch.functional.pad function.
|
Would be nice to merge those extra utility functions @marc-moreaux |
vincentqb
left a comment
vincentqb
left a comment
There was a problem hiding this comment.
Thanks for the PR! There seems to be a significant overlap with existing transforms though, and it's not clear to me which of these transformations would be a non-trivial and useful addition.
| return crop_in_between(tensor, start.item(), end.item(), ch_dim) | ||
|
|
||
|
|
||
| def stretch(tensor, factor, interpolate, ch_dim): |
There was a problem hiding this comment.
Is this similar to Resample or phase_vocoder?
|
Since this PR is out of date and contains significant overlap with existing features, I will close this PR. Please feel free to open a new issue to discuss the transforms you would like to add within torchaudio so we can discuss how to proceed. |
4 participants
In their paper https://arxiv.org/abs/1711.10282, the authors use 2D convolution of monoChannel raw audio sounds with some data augmentation like stretching the sound, randomly cropping a piece of sound and padding the sound on both extremities to allow random crops on the borders