Kokoro is a pretty popular tool- for good reason. Can run on CPUs on desktops and phone. We found it pretty useful ourselves, there being only 1 issue- training custom voices. There was a great tool called KVoiceWalk that solved this. Only 1 problem- it only ran on CPU. Took about 26 hours to train a single voice. So we made significant improvements.
As the name suggests, we added GPU/CUDA support to the tool. Results were 6.5x faster on a 3060. We also created a GUI for easier use, which includes a queuing system for training multiple voices.
Hope this helps the community. We'll be adding this TTS with our own custom voices to our game the coming days. Let me know if you have any questions!
We forked into here- https://github.com/BovineOverlord/kvoicewalk-with-GPU-CUDA-a...
As the name suggests, we added GPU/CUDA support to the tool. Results were 6.5x faster on a 3060. We also created a GUI for easier use, which includes a queuing system for training multiple voices.
Hope this helps the community. We'll be adding this TTS with our own custom voices to our game the coming days. Let me know if you have any questions!