This version adds over 100+ different local voices for all your characters!
Local voices for characters are generated by using the latest text-to-speech technology: VITS.
VITS (Variational Inference with adversarial learning for end-to-end Text-to-Speech) represents a cutting-edge approach in the field of speech synthesis, combining several advanced techniques to generate high-quality, natural-sounding human voices from text. See the paper here.
Layla adds two new apps in this version:
VTCK text to speech
Piper text to speech
These two apps download the required voice packs to use for local offline TTS (text-to-speech). VTCK offers 100+ different voices to choose from. VTCK voices are small to download but comes at a cost of relatively lower quality. Piper, however, offers high quality voices at a cost of a larger download size.
Enabling any of the TTS apps will allow you to choose a voice during character creation!
This update also comes with a new character: KIP! He is a robot created to serve, but has just about had it with humans. This character serves as a technical demo of the local text-to-speech technology.
We plan to add more voice packs, giving you a wide range of different voices to customise your characters with!
Additionally, this is the first step towards offering voice cloning services in the future. We plan to give you the ability to clone any voice and add it to your character, giving Layla a new dimension of personalisation!
Additionally, this update comes with a host of quality-of-life updates and bug fixes.
Changelog
New features:
New app: Piper text-to-speech! - medium quality, larger size, 10+ voices
New app: VCTK text-to-speech! - lower quality, smaller size, lots of voices (100+!)
more voice packs are coming soon!
Improvements:
Goldfish model has been re-trained with significant improvements in cognitive ability
support tablet landscape mode
greeting text in create character screen now supports multi-line
allow creation of characters with empty personality
character background is now set automatically when importing from TavernPNG
added ability to cancel downloading models upon app start
Bug fixes:
fixed bug where custom character greetings were applied twice (this should improve your character's cognition significantly at the start of the convo)
fixed bug where edit chat names was not working on iOS
Note: To-do app has been disabled temporarily due to an incompatibility with TTS
Comments