|
Released preview 8.7.101 this has a completely re-written backend to use the Piper neural TTS voices. It no longer needs a separate process for piper (indeed it doesn't use piper code at all, only the models, and runs the models directly itself). This improves performance/robustness and also implements normalization of audio samples generated from the models which might increase audio clarity for some voices.
I'd appreciate some testing to ensure I didnt break anything too badly.
|