Batch #28 - 15 (6+9 bonus) new voices; +1 game; 400+ voices! ; short break to prepare release

The next batch is here! A fairly big one, as there was a spike in requests (one of which private). The voices from the poll are:

- Fallout 3: MaleAdult06
- Fallout 3: Butch
- Cyberpunk: Rhino
- Cyberpunk: Kerry
- Borderlands: Claptrap
- Borderlands: Jack

The voices from the public requests are:

- Fallout3/4/New Vegas/76: Narrator (aka Newscaster for F4)
- Final Fantasy: Aerith
- Final Fantasy: Cloud
- Hyperdimension: Cave
- Hyperdimension: Compa
- Hyperdimension: Nepgear
- Hyperdimension: Neptune
- Hyperdimension: Noire
- Hyperdimension: Vert

This includes voices for a newly supported game series, Hyperdimension! I've included the asset file for this below, place it in the /resources/app/assets folder, making sure it's called: hyperdimension-e29eef-HD-Hyperdimension.jpg

The Narrator/Newscaster voice is present in all 4 Fallout games, but it's the same voice, so any of them will do, though I've included versions for all games, depending on preference.

--

With this batch, we've now crossed the 400 voices milestone! We'll need to have a party or something when we cross 500. On that note, I will be skipping the next poll for now (apart from some remaining requests), as I definitely need to spend as much time as possible preparing the v2 release, so I can get this out. You'll be happy to know that the actual release is finished! The holdup at the moment is with training voice models using the new FastPitch 1.1 architecture. 

For some reason, it's seemingly quite difficult to converge a model with high quality, on voices other than the LJSpeech voice used for research benchmarking. This new model has some important new features necessary for a number of features in the next release, so it's critical that these models work well. The app will be out once there are at least a small handful of these new models trained to a sufficient standard.

The plan for xVASynth voices (once I get past the training issue) is to go back through all (400!) voices trained so far, and re-train them using this new model architecture. That sounds like quite a feat now, given the backlog, but I'm hoping/expecting that RE-training will be considerably quicker - we'll see. If not, I'll also be switching over to focusing development efforts on xVATrainer, and we could re-do them together, for those with the hardware required for voice training.

Stay tuned for updates!