the great utau crashout 2: i'm chill
so... i had a bit of an episode last blog post :)
well, after i got some rest at an unreasonable hour after spending the back half of the night typing, i remembered i still had a helpline available to resolve my conundrum: chevrefee themself, creator and manager of all the english c+v tutorials i was using.
this was a last resort and i didn't expect them to respond, and i was also getting used to the idea of just regenerating my samples from scratch and doing everything all over again.
thankfully, they responded and we got in touch, and my issues lied in something i would have never thought to check: the sampling rate for clips that were downloaded straight off of talkmodachi, much like my haphazard ey replacement clip. apparently openutau flat-out rejects clips that are not a certain sample rate, which is why certain clips that i didn't do extra editing (in extending their beginning silence with a DAW) were insufficient and silenced without much comment.
on one hand, i am a little flustered i belabored my desperation in my help message to this saint of a creator only for it to be a discrepancy easily found in looking at file properties. but on the other... i would have never thought of this!!!! if i started from scratch and decided to not to give my samples the same treatment in the DAW, i would have ended up with a voicebank that couldn't sing at all. which i thing would have thrown me over the edge. i can't thank chevrefee enough for being patient with me TwT
the state of kandyloid now is that she is definitely workable with the limitations of her entire setup. her y sounds are very harsh as they are common in the lyrics i make her sing, though i was able to smoothly work around this with some file editing ;) and i'm proud! she sounds so much better without sounding phlegmy every ten seconds!
there is still work to be done with editing. some consonants are still harsh, plus i think i fucked up oto'ing some syllables because their transitions look WACK. but i can rest easy now that i know 1) how to get the samples to work (the key is to not use them fresh out of talkmodachi that's for sure!) and 2) if one were to look past her fiddly qualities, i can get a really cute little voicebank after this!
after resolving my issue with the files (and rexporting everything to make sure every single sample had consistent start times, channels, and SAMPLE RATE), i got kandy to sing daisy bell again PROPERLY and, while it's not quite the same as her first pass (i completely redid all of the vowels during this rebuild process), i felt emotional all over again hearing her speak, speak in a voice that finally feels right for her...!!!!
the next step, after i clean up the english cv to the best of my ability, is to venture into making a japanese cv again using talkmodachi samples. it looks a little more involved and might take some extra effort to make the us english mii voice pronounce japanese phonemes correctly, but tbh...? workable. i can do that....
while i shouldn't get ahead of myself just yet, i am looking forward to the next step of kandyloid, despite the lowest of the lows i reached previously :) i have learned and am learning!!!!!!!!!!!!!!!!!!!!!