andabi / music-source-separation

Deep neural networks for separating singing voice from music written in TensorFlow
795 stars 150 forks source link

Possible bug on bss_eval_global #34

Open tensorfreitas opened 5 years ago

tensorfreitas commented 5 years ago

Hi,

First of thanks for sharing the code and all the insights. I was playing around with your code and noticed a possible issue on eval.py, more specifically on bss_eval_global.

When debugging in this function I noticed that when you define gnsdr = gsir = gsar = np.zeros(2), you are linking all those variables to the same array. Therefore, every time gnsdr is changed, gsir and gsar are also changed accordingly. I suppose this is not what was intended, and correcting it to

gnsdr = np.zeros(2) gsir = np.zeros(2) gsar = np.zeros(2)

Would solve this problem.

Best Regards