so! im throwing these all in here bc theyre not gonna be shared outside of just. dumping as tests. been fucking around a lot with the vocalo change to test its limits and ill give a quick summary of what i determined:
- clean vocals, clarity with what is being said, and an accent that won't skew the interpretation is important. all my tests were with vocals ripped using AI, which i thought were very clean, but even so their quality and even minimal bg noise wasn't good for v6! regarding accent, i didnt have too much of an issue but for example, i tried vocalo change over a singer like marina and the diamons, who heavily enunciates and exaggerates, and v6 had some odd results with how it took it, interpreting the found phonetic or missing consonants entirely.
- it will be exact- in the sense that if you use it on another synth voice you rendered, it wont really.. improve the pronunciation (example: testing this on fukase's accented english resulted in, well. allen singing accented english).
- sometimes is exactness on "unqiue" singers kinda makes it hard for me to feel like its actually changing much at all- perhaps im too familiar with some of the samples to not hear the original, idk. but i rendered a lot of these with Akito because i think it was *less* obvious when using the opposite gender voice. female voice on female voice, or male on male, while in their range will make them blend a lot for better or worse (mostly if your vocal quality isnt good).
- im sure most would know this but. its important you bring voices down an octave for the males if theyre too high lol.
ok thats enough. heres all the random things i did -
note that some of these sound bad. thats why i am sharing them here lol. OH and if youre thinking 'wow thats a lot of musical songs', its bc with all these being AI ripped vocals, musical voices often come out the most clean, so they were the best examples out of others i had lol. in the end i think this gives a good idea that no, you cannot get a super clean cover with AI ripped vocals unless its a very empty song and very high quality file.
Waiting in the Wings (tangled) with
Akito
Listen to a track by morrysillusion.
www.whyp.it
the original voice is quite unique with belting, and while these vocals were not the most clean (mostly at the end where more instruments came in), it still struggled with higher sounds in places where it was otherwise fine.
Just Your Problem (adventure time) with
Akito (with talking parts included)
Listen to a track by morrysillusion.
www.whyp.it
this was ripped from a youtube upload, out of curiosity on quality. it loves to warble. but like, not terrible. but i think its still very obvious, even with clean up, that this was doen off a low quality file, it misses some consonants those breaths are bad (i cut out some ear piercing breaths), as always a struggle with louder high parts.
Liquid Smooth (mitski) with
Akito (short because the rest did not work lol)
Listen to a track by morrysillusion.
www.whyp.it
a test with a busier, non-musical song. lots of issues, and i dont even give you the full thing because it *did not* work lol. but theres a bit of japanese in here too so its cool to hear how well he does both languages, i think it becomes more obvious its off of Akito when that happens.
Rather Be Me (steven universe) with
Akito
Listen to a track by morrysillusion.
www.whyp.it
arguably... the cleanest one, which again is ripped vocals.
Its Over Isnt It (steven universe) with
Sarah
Listen to a track by morrysillusion.
www.whyp.it
i was doing this with solaria, and the vocals from the original were very very clean. im kinda shocked this still worked so well, especially on the belting. i mentioned how i feel you get some of the original singer when you use the same gender voice, and Dee dee's singing is p obvious to me in parts, usually the lower parts? that maybe Sarah isnt really built for? idk.
Barbie Princess and the Pauper dialogue with
Akito (Erika) &
Allen (Anneliese)
Listen to a track by morrysillusion.
www.whyp.it
i mentioned this in a status. it was mostly a joke when i put them on the ripped vocals i had for this song... and uh. well, idk. its hilarious to hear how they work on talking.
thaaaats it. all the tests worth showing. more out of curiosity and to show yall how this sounds on vocals that are not clean stems. i do think thatll deter people from wanting to recover every song in existence by slapping them on vocalo change- which is good, bc its only really cool to understand how it works. its very clearly best with clean recorded vocals (many of these had reverb of some kind and it really messes things further), which is how its advertised.