Hi everyone! Using my tiny fleet of GTX 1060’s, I’ve trained two new voices using Larynx. This brings us to:
All of these voices were trained on public speech data, two from folks on the MozillaTTS forum. They’re available as Docker images for Raspberry Pi 2-4 and PC, as well as Hass.io add-ons. Each Docker image can emulate the MaryTTS API, so you can use them from Rhasspy, Home Assistant, or whatever their native MaryTTS plug-ins.
In parallel with the training of the voices above, we have Rhasspy community members both recording their own voices and helping me develop sets of phonetically rich sentences.
These contributions will give everyone a variety of voices to choose from, but will also help ensure the voices sound right to native speakers. Minor errors have already been noticed for the Dutch and German voices above, possibly because there weren’t enough examples of some phoneme pairs or triples (or I’m doing something wrong). Either way, things should only get better with time