Just found this project on Github that converts many open source TTS models to run as native TTS engines for android. All prepackaged so you don't have to write any code or anything. I don't use android myself, but I'll paste the link here if you wanna check it out. I'd be really curious to see how responsive some of these are. The latest Kokoro model is up there near the top, along with a lot of voices built using Piper. k2-fsa.github.io/sherpa/onnx/t…

Tamas G reshared this.

in reply to valiant8086

@valiant8086 Wow, well I'll say aside from responsiveness, the actual voice quality is really good, some of the best you're likely to get out of open source models. Likely the reason it's so sluggish is the model size, Kokoro is about 82000000 parameters which I thought could be small enough for good speed with a screen reader but I guess I was wrong. I'm sure some optimizations could be made to the code to get the model to run better. Can confirm that voice is Adam from Eleven Labs.