results reproducibility

[taras-sereda]

Hey! I found your implementation of TasNet, also working on it.
Wanted to ask few questions, share my observations.

1. Were you able to reproduce paper results?
2. In my experiments I don't see too much change in source masks over training, also I observed that LSTM cells becoming really high in values, and clipping is necessary to achieve stability.

Thanks in advance, will be happy to discuss this paper and share results when I'll finally be able to reproduce them.