ONNX

Hey! I’ve been working on another fine-tune of my Czech Piper voice, Kasandra.

This version focuses much more on natural prosody, emotional delivery, emphasis, and sentence endings. The biggest improvement isn’t really noticeable in short neutral sentences, but in longer speech.

What surprised me is that she can now reproduce the emotional delivery from the training recordings surprisingly well. Honestly, there are some lines where I don’t think I could read them with the same emotional expression myself. 😄

It’s getting pretty close to an ElevenLabs-style voice, but running locally with Piper.

Ready to merge
This branch is ready to get merged automatically.

Sign up or log in to comment