Add JSON API.

[jonashaag]

Also optimise Dockerfile for quick rebuilds.

Also add multiple input files support to predict.py.

Is it expected that the estimate for 5 speakers example is 6?

[jonashaag]

Also, what is the purpose of predict_audio.py?

[faroit]

@jonashaag as far as I remember I got 6 with the tf backend and 5 with theano. But I will check later.

[faroit]

@jonashaag can be removed

[faroit]

Is it expected that the estimate for 5 speakers example is 6?

i checked again, for both CRNN and RNN i get the correct response (5). So you probably have fallen into the wrong tf backend.

[jonashaag]

OK, changed to Theano backend. I also switched to Conda so that any numeric libraries (such as MKL) are automatically installed

[faroit]

Okay. Great. I will probably add a simple unit test to make sure that the backends are correct