Maybe a bug when duration prediction output 0 length #22
Description
My training crashed on step: rfwave/dit.py, line377, in forward
x_ref_fill[i, :, ctx_start[i] + ctx_length[i]] = x_ref[i, :, : ctx_length[i]]
RuntimeError: The expanded size of the tensor (0) must match the existing size (201) at non-singuleton dimension 1, Target size: [512, 0], Tensor sizes: [512, 201]
This happens at validation step, so I think the ctx_length predicted by duration predictor may get zero value. is there any such risk?