#What kind of sounds are bad to use when training your voice model?

1 messages · Page 1 of 1 (latest)

keen obsidian
#

Hey guys! I'm trying to learn how to create voice models with the highest quality, but I'm having difficulty understanding which types of audio are bad to use.

Questions:

  1. Are the sounds of crying, panting, laughing and screaming bad? Or would having these sounds help the AI ​​understand these more specific tones to replicate?
  2. Is there a problem with using different tones of voice, even if it is the same person? (The same voice actor/voice actress making different tones for different characters).

Samples of what I'm talking about:

#

If these samples from the first audio are used in training, even if they are sounds from the character himself, could this somehow harm the final quality? Would the AI ​​interpret the sound of panting and breathing as part of the speaking voice? (Speaking while blowing).

And the samples from the second audio, if there are different tones of the same voice (There are 3 in the sample audio), would this harm the final quality, with the AI ​​trying to create a single voice that is halfway between all the tones or would the training allow the model to perform all the voice tones that were trained, needing only to change the pitch? (Change the pitch to the tone that most resembles character X or Y of the same voice).

lethal crow
#

no, these sounds aren't bad
the only type of bad sounds for datasets are any type of sound that was not made by a human being

#

bad sounds are:
reverb
room reverb
non human made sounds

#

and yeah, its aware panting and breathing are features of the dataset and learns it

#

using different tones is bad yeah

#

too much diversity (like when someone does a character impression) confuses rvc

lethal crow