Z Idol Talker

Z Idol Talker is a standalone application for running Piper TTS models. For information on Piper please check GitHub: https://github.com/rhasspy/piper

This application was made in about a week as a proof on concept. While there are a few more things I will work on over the next month, this program will not get frequent updates unless there is a demand for it. Nonetheless, feel free to give suggestions on how I can make the program better.

This program is not to be used with voicebanks trained on samples you don't own the rights to, nor should it be used in ways that violate public morals.

Patch Notes can be viewed here: https://www.zvocalproject.com/z-idol-talker-patch-notes/

Features:

• Easy Voicebank Management: Voicebanks within your set voicebank folder will appear in an easy to navigate list, and can be loaded by clicking on their icon. Voicebanks appear alongside their name, language, voicetype, icon, company name, and a button to easily open their license file.
  
• Fully Customizable UI: Each section of the UI can be replaced with an image that is loaded alongside each voicebank*. (These examples are a bit of a rush job lol)
  • Defoko English is an easy jumping off point as she has everything customized (Aside from the top row which will be rectified soon). An example voicebank with the default UI assets will also be available shortly.
  • *The voicebank selection assets associated with each voicebank are loaded based on the voicebank they represent as opposed to the active voicebank.
• Mystery Button: I wonder what it does?

Voicebanks:

The download page includes non-commercial voicebanks of Z Vocal Project characters and also Defoko English. These models were made to test the engine and were only trained for about an hour, leading to a very mixed quality. Defoko will receive an update with more data in the near future, while the others depend entirely on demand. These voicebanks are:

• Defoko English, Dunder Announcer, Dunder Castilian, Dunder Spanish (Latin), Dunder French, Gnomeder, Ascenia, Dunder English, Ethel, Ambrose

Additionally, I have prepared a whole slew of other models in Z Idol Talker format. The password for each page is "ZIdolTalker" As they were a bit much to manage, all extra voicebanks aside from Tsukuyomi-Chan English have been delisted, the password is still ZIdolTalker

• "Original" voicebank Tsukuyomi-Chan English:  https://dunderhyphenp.itch.io/z-idol-talker-voicebanks

Supported Languages:

Arabic, Catalan, Czech, Danish,  German, Greek,  English,  Spanish,  Finnish, French,  Hungarian, Icelandic,  Italian, Georgian,  Kazakh, Luxembourgish, Nepali,  Dutch, Norwegian,  Polish, Portuguese, Romanian, Russian, Serbian, Swedish, Swahili, Turkish, Ukrainian, Vietnamese, Chinese

I am working on getting Japanese working but I either need a much larger corpus, or the way espeak handles Japanese is completely broken. Currently your best bet is to train a model on Spanish and alter the language/espeak codes in the JSON file to" ja". Even then you will be limited to hiragana inputs and it might be more worth your time to just input romaji into the Spanish model.

Coming soon:

Over the next month I am going to work on text and video resources for using and making voicebanks for Z Idol Talker/Piper.

As for the program itself I am working on making the accessibility mode work a bit better and better UI scaling.