Documentation update

[oadams]

Things to document:

• The need to build Kaldi for filterbank_and_pitch features.
• Need for settings.ini / include a default.
• Model training arguments.
• Use of train_prefixes.txt, valid_prefixes.txt, test_prefixes.txt
• Mention Python 3.7 is not supported by Tensorflow.
• Point users to Github issue tracker to raise problems.
• Document assumptions about data, and how the tool is primarily aimed at the single-speaker scenario.

[shuttle1987]

Is there any test data for filterbank_and_pitches? I'm putting together environment build automation over in PR #208 and I'd like to be able to verify that I can automate the install of Kaldi by running some data through these code paths.

[oadams]

There's no data: the distinction between filterbank and filterbank_and_pitches is only that pitch features are extracted from the audio additionally, so a test for pitch extraction could use the same data as any existing test that extracts filterbank features.

[shuttle1987]

In that case perhaps we can just use an existing test set for the data but with the different parameters for this? I'd like to get some test that requires Kaldi to work

[oadams]

Yes, that's right.

[shuttle1987]

Is this issue related to #86 and #1 ?

[oadams]

Yeah, perhaps we should just enumerate current documentation change requirements in the checklist above, unless its really big enough and requires implementation that somehow warrants its own issue.