Is the methodology of getting source and dest points logical in Time Warp?

[PrathamSoni]

`def time_warp(spec, W=5):
num_rows = spec.shape[1]
spec_len = spec.shape[2]
device = spec.device
y = num_rows//2
horizontal_line_at_ctr = spec[0][y]
assert len(horizontal_line_at_ctr) == spec_len

point_to_warp = horizontal_line_at_ctr[random.randrange(W, spec_len - W)]
assert isinstance(point_to_warp, torch.Tensor)

# Uniform distribution from (0,W) with chance to be up to W negative
dist_to_warp = random.randrange(-W, W)
src_pts, dest_pts = (torch.tensor([[[y, point_to_warp]]], device=device), 
                     torch.tensor([[[y, point_to_warp + dist_to_warp]]], device=device))
warped_spectro, dense_flows = sparse_image_warp(spec, src_pts, dest_pts)
return warped_spectro.squeeze(3)`

Based on my understanding, the point given should be a time coordinate. Currently, the point to wrap is returning an actual value from the spectrogram, which does not align with the computation of source and dest points.