Add more machine learning backends

[pzinemanas]

pytorch and scikit-learn are good candidates.

[elcronos]

Hi @pzinemanas. I'm a PhD student and like me, many in the academia has a preference for Pytorch. However, in terms of Environmental Sound Classification models, the majority of source code available is implemented on tensorflow/keras. I reckon a pytorch version of this library would be a great contribution !

[pzinemanas]

Hi @elcronos , thank you very much for using the library and giving your valuable feedback. That's true, most of the researchers are now using pytorch. We are working on this (see for instance #24).