Speaker manager for multi-speaker handling

[erogol]

• SpeakerManager to interface multi-speaker models.
• Use SpeakerManager in Synthesizer
• Adapt SC-Glow models from @Edresson using SpeakerManager .
• Tests for SpeakerManager
• Update demo server to use multi-speaker models.
• Update synthesize.py to use multi-speaker models.
• Update training scripts to use SpeakerManager in multi-speaker case.

Enabling multi-speaker models means;

• Let people list speaker of a multi-speaker model.
• Choose a certain speaker to generate voice
• Let people upload voice samples to mimic the multi-speaker model.

To call the multi-speaker models with this PR;

$ tts --model_name tts_models/en/vctk/sc-glow-tts --list_models  # list the available speakers for the model
$ tts --text "This is a multi-speaker model.." --out_path output.wav  --model_name tts_models/en/vctk/sc-glow-tts --speaker_idx 'p333'     # run the model
$ tts-server --model_name tts_models/en/vctk/sc-glow-tts  # run the demo server 

[Edresson]

Apparently, everything is ok, great addition is much better visually than the old version :)

[kirianguiller]

Hi @erogol , you had any trouble for including the #5 in it ?

[erogol]

@kirianguiller what do you mean?

I started with your PR. Rebased it and wrote a bunch more :).