#Piper voices with better pacing
1 messages · Page 1 of 1 (latest)
What goes in, so goes out. Most likely, the datasets for the voices you're listening to didn't include all the punctuation marks. It's important to focus not so much on punctuation rules, but on the actual pronunciation of phrases. Also, Piper has a fairly limited number of intonation tokens; in addition to ",.!?", it uses a colon and an em dash.
If a speaker uses inconsistent intonation, even a large training batch won't eliminate strange pauses in the final result. At least, this is true for datasets, lasting a few hours or less.
terrible input. piper DOES support various runtime props, including speed, and from my finding the addon has support for setting these values.
you're looking to set the length_scale value, between 0.0 to 1.0