for testing can we use smaller batch size?

[Benuraj]

is it a necessity to put --batch_size as 100 for test clips? Why lower batch crashes the code? What this --batch_size 100 represents while running inference?

[Mauker1]

If I recall correctly, you have to use the same batch size that was used during training, as the other parameters.

The batch size indicates how many seconds of audio will SEGAN process at each time.

I've successfuly trained it using a batch size of 70 seconds with my GeForce 1060ti with 6GB of dedicated memory.