Open crazygirl1992 opened 6 years ago
If you are training overlapped speakers, the answer is no.
There is a small bug in this project.
Suppose you set the validation sample number is 50 while you run python prepare.py ...
. A file called Thumbs.db may be put in the val directory and be counted as one of the val data. So in fact you got 49 validation samples. The validation_step is then be calculate as 0, which causes an error.
A simple solution is that you move one sample in the train directory to the val directory.
whether the validation_steps need to set by myself?