#Making my Serbian more Native English with my AI voice clone?

1 messages · Page 1 of 1 (latest)

foggy nacelle
#

Hey,

So I wanna clone my own voice but my English pronunciation isn’t that great.
I cloned my voice few times but when i make a cover for example, you can hear that my pronunciation is a bit off.

What would be the best approach to this?

Which PreTrains, whichc models, what settings etc

Thanks ❤️

jagged viper
#

dont use an index, try pronouncing english characters properly

#

you dont the entire dataset of you speaking english, but at least some proper pronunciation for what you notice is off

foggy nacelle
foggy nacelle
#

❤️

jagged viper
#

the only fix for that is to have enough correct 'the's in the dataset.

foggy nacelle
#

ohh i see, understood ! Thank you, and do you maybe know some tips for how to lose that "SSS" after inference generally with every model?

jagged viper
#

of you're getting some horrible noise instead of silence, it is likely you have either trained a model or using 24-bit audio for inference. Run a basic denoiser.

foggy nacelle
#

denoiser for dataset? you know what i mean? that typical AI SSS on sibilants?

jagged viper
#

oh, you mean weird S sounds, not the horrible noise instead of silence

#

having more data may help

foggy nacelle
#

yupp

jagged viper
#

it is a big problem for rvc as sibilants are made out of white noise

foggy nacelle
#

bro u made my day btw with "no index" solution, i ve been doing this AI for years but never tried that ...

#

thank you

jagged viper
#

you can also take an index from an english model

foggy nacelle
#

oh so i just download few random male models singers and i just put it in a zip with my model?

jagged viper
#

i mean for your own inference

foggy nacelle
#

could you give an example?

foggy nacelle
jagged viper
#

I mean dont use a random index to upload it to weights.gg

#

i mean use it to run inference

foggy nacelle
#

got it ❤️