Voicebank UTAU Development thread - unnamed 100% Artificial voicebank

[morrysillusion]

i just made a status about this earlier and decided, why not? i want to genuinely try this so i felt like making a thread for it would be best. this OP post will be updated to contain any downloads for the voicebank (and the thread title will change to whenever it is released and has a name).

at the moment my knowledge on doing this is pretty limited, please chime in with information on how to create artificial sounds as it would be very very useful! especially so i can improve it!

My rules/what i consider valid for this bank:

• sound must be completely computer generated with no human or non-digital source involved
• VSTs/plugins/effects are allowed to manipulate the sample (using a 'gender changer', phaser, distortion, etc to make the final samples more unique)
• no AI to train or enhance the voice

Programs i am using:

• Praat - for vocal manipulation and sound generation
• audacity for manipulating final sample tones/sound
• FL studio for some other editing and other effects

Goals for the voice:

• aiming for something feminine but bordering on androgynous
• CV so its robotic, but with good clarity for understanding it
• maybe arpasing/english at some point

Most recent sample can be heard here:

last updated October 19th 2024
notes: changed tone to be less soft. created almost all sounds mainly to test vowels. "i" and "o" are struggling to sound different, consonants themself are not full edited to have the same effects of the vowels so theyre not perfect

 

[morrysillusion]

update! made some changes to the voice tone to make it less whispery. reminder by no means was the first test supposed to just stay that way. this is gonna have a lot of experimenting that may result in some pretty different versions.

sample_ghostrule_pinkalpha1

Listen to a track by morrysillusion.

whyp.it

(the sample i link doesnt give a huge picture of the status the bank is in right now)

most sounds for CV were made mainly so i could test the potential sound quality. even though the samples are easier to distinguish outside of utau, the "i" sound keeps coming out really similar to "o" and so im trying to figure that out before i get any further on adjusting tone... im not entirely sure how to keep tweaking it however... it doesnt seem obvious when theyre not near each other, but when they play one after the other multiple times it is way more muddy

right now the consonants are technically weird if you were to take a look at the samples. 1) i could only really figure them out by splicing them together with the vowels which makes it weird in many places 2) i havent put the "effects" i had put on the vowels to make them match, even tho it may not be that needed it is still a bit apparent in some sounds. in the end i dont really need the samples to come together perfectly, just really need to aim for the vowels being distinct, and enough consonants being distinct as well that you can tell something is being said and its not just noise.