Bottom text.
#What kind of result do you get if you train a dataset with samples from two different voices?
cerulean bobcat
whole bone
when you mix multiple voices in the same dataset you model may randomply switch voices based on the audio it tries to finder to provide a better match for the pitch.
cerulean bobcat
Oh, so you wouldn't get a voice that sounds like their timbre is somewhere in-between the two original voices?
whole bone
nope
the model would fall back on the best match it has
so if you have a male and female voices mixed, it may speak with male voice but sing with female
wise pivot
Oh, so you wouldn't get a voice that sounds like their timbre is somewhere in-be...
train both datasets separately, first the male dataset, then the female dataset
after that merge both models with a value of 0.5
the result will be unique voice that have the characteristics of the original voices
it'll be good? hard to tell, we can't predict that
cerulean bobcat
Maybe it would work better with two male or two female voices?
Or voices that just sound more similar to each other
cerulean bobcat
👀