I noticed that Solaria V2 pronounces the English word “me” as “m eh” instead of “m iy”
I have the standard English dictionary loaded and the song is also English. Does anyone know what the issue is?
1 Like
I have similar concern so I’m adding on this thread. I am using Yuma 2 in English, and the vowels are pronounced weirdly… I’m trying to input the word “you” – though it shows “y uw” the vocal is pronouncing it more like “y uh” or even a schwa. Adjusting vocal mode pronunciation doesn’t work to fix this either.
With testing my V2 voices so far, V2 vowel sounds in English (especially in cross-language synthesis) feels weird. It doesn’t feel as pure or straightforward as V1, but more stylized to some kind of American pop accent in some aspects, or whatever the accent is. I am not quite a fan of how the vowels sound right now, it just doesn’t “sound right”
1 Like
These unusual pronunciations from Solaria, coupled with the lack of global vibrato control has left my V2 as totally unworkable. Hopefully Dreamtonics get to grips with the voice problems, along with the multitude of software issues sooner rather than later.
2 Likes
Solaria V2 doesn’t sound at all like a native English voice. It sounds more like an asian voice. For me, three things have happened here: it’s a bug, they tried to modify Solaria to better adapt to languages like Chinese, Japanese, etc., or they used an asian singer with a similar voice to modify V1 and turn it into V2.
Solaria even knows V2 is meh. 
I agree a bit about the pronounciations. There are now several banks which have suddenly noticeable accent, when previously not. It is like picking Japanese or Spanish a bit; like a bug selecting phonemes from X-SAMPA instead of ARPABET. Also some Spanish people are saying that now for example Saros has English accent, so the other way around. Also dictionaries produce some funny results now
Yep, accents and enunciation is clearly worse in V2.
Comparing the same song on both V1 and V2, adjusting independently, V1 is way more predictable, stable and pronounces better.
V2 is capable of more interesting and lively interpretations, but it needs a lot of attempts before it gets a good take. It has worst pronunciation, and pitch is unstable, with a lot of really weird pitch wobbling and strange artifacts. Then, when it gives you a good take, it is way better than V1, but it takes a lot of effort or random luck to get a good one.
V1 is like a good student that always gets the job done, but lacks that special talent that makes a star. V2 is a highly talented person with the potential to be a star, but that hasn’t even started to train in his craft, so he doesn’t know what he is doing even when he gets it right.
Trying to use Liam and I gave up stringing the words “and fall” together. It creates a long stutter on the “f” sound. Only way to get rid of the stutter is delete the “d” but then it becomes “an fall”.
Maybe I’m not understanding the new phoneme panel, but sliding the “d” around just delays it, I can’t seem to separate it from the 'f" which I guess is creating the stutter. Did not have this problem in V1 with Jin.
I’m experiencing pronunciation problems with Ninezero as well. And his accent is way more pronounced and evident now. Lyrics that used to work fine in v1, now sound all wrong and needing lots of tweaking to get less worse, not good.
1 Like
Can’t say I’ve been able to replicate that… no matter how far I push the diphthongs into each other… Are you on the latest build?
Thanks for testing that out. I had tried everything including resetting phonemes but I went back to it and tried resetting phonemes AND timing, that cleared the glitch. Very strange as my manual results and the reset result are identical to the eye but there must have been some glitching that it couldn’t get past in that instance before the full reset.
Well, you know the old adage “Turn if off and start again !”
For my part I had an issue with the word “clear” with Solaria II.
In my lyrics I have a line ending with “clear” and the next with “near”, where the “ear” part should sound the same.
They get translated to these phonemes respectively: “k l ih r” and “n ih r” so they should indeed sound the same. While “near” was correct, “clear” sounded more like “bear” (b eh r) and sounded like “k l eh r” instead.
I was working on other aspects and wanted to come back to this later, and I did some AI retakes for other purposes, and then the pronunciation for “clear” changed and became correct.
I have issues with Solaria I and Solaria II with ‘for’ Comes out as a fourrrrr. I can improve it by changing the word to ‘four’ and messing with the ‘r’ phenome but it is a regular one.
Odd thing is I can have the word twice within a couple of bars, and it will sound different each time.
I’ve had some similar issues with Solaria - particularly with ‘fall’ ‘crawl’ etc. The solution I found was to use the new ‘mouth opening’ parameter to adjust the sound of the phoneme. Works well for me, maybe give that a try.
One other pronunciation issue I’m having - and am interested in hearing if others experience the same - is that often the sibilance at a start of a word sounds HEAVILY de-essed. Almost like there’s a lisp there instead of a straight ‘s’ sound. It’s driving me up the wall!
I’m getting “end” instead of “and” with Solaria 2. The phoneme for the A is set to ax. I changed it to ae and the pronunciation was corrected. I’ve noticed other Synth V 1 voices doing this previously but Solaria 1 always had better pronunciation.
There is another one I’ve run into yesterday, but it is more an issue of translating a word into the correct phoneme. The word “the” can be pronounced differently depending on the next word, for instance “the cat” vs “the other”
In my song, whenever i try to use “the tears”, it uses the wrong “the” as if the next word started with a vowel. In this case i have to manually type the phonemes of the correct “the” version.
Again, seems like a word to phoneme issue, the second syllable of “appart” gets translated to “p er t” which doesn’t sound right (sounds like “appert” instead of “appart”). I added a dictionary entry to make it use the correct phoneme. At least this is easy to fix.
Interestingly, making a song entirely in Latin by manually writing spanish phonemes is going much smoother than anticipated.