When using a 3-second model to generate a word, the speech effect is not correct.
#1
by taonjfu - opened
First of all, this project is great — the speed and memory usage are both excellent. However, when generating short phrases (under about 1 second or around 1 second use voice except heart), the speech quality is not so good. I'm not sure whether it's an issue with how I'm using it or a problem with the model itself.
thanks man!
I am tweaking the model, the tokenizer and phonemizer a bunch aright now for another project (https://gist.si) I am working on. Your best bet is to use the sister Github repo to generate a newer version yourself https://github.com/mattmireles/kokoro-coreml
Once I am done jamming on gist, I will update this project with a stable newer, working version of the mode.
LMK if this helps.