Add synthesizer preprocessing support for other datasets

[ghost]

Resolves #413 by adding preprocessing support for other datasets. This is designed with LibriTTS in mind but is compatible with other datasets (e.g. VCTK) by using the same directory structure and file naming convention. This change is designed to be almost transparent for those using LibriSpeech. The training documentation on the wiki requires no update.

These changes have been tested to ensure both LibriSpeech and LibriTTS can be processed and used in training without errors.

Changes

1. Remove hardcoding of dataset folder name and subfolders in synthesizer/preprocess.py. These are now specified with new command line arguments:
   • --dataset_name (defaults to "LibriSpeech")
   • --subfolders (defaults to train-clean-100, train-clean-360)
2. New command line argument "--no_alignments" (default False)
   • If True, does not split audio clips. For a given wave file "filename.ext" the preprocessing script will look for the text in "filename.txt" or "filename.normalized.txt" in the same location.

[mbdash]

@blue-fish, Would it be useful if I was to offer a GPU (2080 ti) for contributing on training a new model based on LibriTTS ?
I have yet to train any models and would gladly exchange GPU time for an opportunity to learn.
I wonder how long it would take on a single 2080 ti.