训练时将segment_seconds改为0.1,hop_seconds改为0.01,frames_per_second改为20出现结果为tensor(nan, grad_fn=<AddBackward0>)
sheng-zhong opened this issue · 1 comments
sheng-zhong commented
训练时将segment_seconds改为0.1,hop_seconds改为0.01,frames_per_second改为20出现结果为tensor(nan, grad_fn=)
lzh88998 commented
这个样本是不是太小了,容易产生singular矩阵?