Skip to content

v1.2.0-beta.1

Pre-release
Pre-release

Choose a tag to compare

@liisaratsep liisaratsep released this 16 Oct 08:15
· 1 commit to multi-style since this release

A version with style control and transfer options. This is the first version and still somewhat rough. Inference can be used via the code in Koodivaramu. The model supports 10 speakers and 3 styles, however, it was trained with data from more than 700 speakers to help with pronunciation robustness and versatility. The main benefit of this model is more natural prosody with longer conversational texts. For production use, we still recommend using a multi-speaker model for reliability and wait for a more stable release.

Model: https://huggingface.co/tartuNLP/fastpitch-est-multistyle