Few changes regarding the synthesizing of the vocoder training data. Uploaded trained models of the vocoder to the git.

Modified the synthesizer preprocessor to write unmodified audio to the disk. Refactored a lot of code in wavernn. Adapted wavernn to handle a mu-law

Reverted relative imports, separated the encoder package and the encoder scripts instead. Improved on the umap demo

Renamed SV2TTS to encoder. Finished basic interfacing of the encoder. Changed all imports of the encoder to be relative imports in order to use it as a package