I’m starting to have a look at the differences (improvements ?) between SV1 offerings and SV2…
Here’s the first (short comparison of Mai 1 vs Mai 2)
It’s interesting to note that I resisted the urge to type in the lyrics, but let YouTube have a go to see how accurate it would be - the only mistake was ‘outing’ taken to be ‘now think’. The previous word exhibition ends in n so maybe there’s too much liasion between this and the o of outing ???
Second short comparison with Mai, this time changing the timbre for a slightly more mature sound…
Once again I resisted the temptation to type in the lyrics to see what YouTube’s algorithm came up with… Only mistake is with the Japanese pronunciation of ‘Mai’.
Third short comparison with Mai, this time much more ‘male’ timbre dialed in and I’m starting to explore the Open-mouth and Phoneme offerings - got a long way to go to get these just right !
NOTE: The default settings in SV2 for Mai offer dx instead of t in the word to and the ‘r’ of really is so lame and lispy that it requires major adjustment - I’ve spend enough time trying to get a good sound that I’ve posted as is for now - Enjoy !
There’s no manual at the time of writing, so guess work and observational listening is paramount.
This is how I think the phoneme triangular thing works, at least that’s what my ears tell me…
No difficulty with SV1 Mai and YouTube Algorithm, SV2, however threw up “sinking” instead of singing and “thir” instead of third…
Forth short comparison with Mai, ‘out the box’ vocals just to see where the difficulties lie with Mai 2…
Once again I let YouTube auto-generate subtitles and guess what? SV1 had no problem, but SV2 did - the recognised as that and hoping recognised as helping.
There is definitely something wrong with the way vowels are pronounced. Whether it’s the mouth opening thing not set correctly or just sloppy work to get the program out the door…
@Bobox, thanks for these. I’m enjoying them. One small bit of (unsolicited) feedback is the vids tend to have 7-10 seconds of silence before beginning and it can seem like an eternity to get the magic snippet from such a short vid. Small request for less (or no) silence before starting the content. Regardless, I really appreciate the time and effort you’re giving to put these out there!
Hi, Thanks for the suggestion, I’ve added a start time to the videos that effectively skips the intros…
If anyone else wants to know how to do this:
- If your YouTube video link is in the format …youtu.be/… add ?t=X at the end (Where X is the number of seconds to skip)
- If your YouTube video link is in the format …youtube.com/watch?v=… add &t=Xs (Where X is the number of seconds to skip)
A scholar and a gentleman! 
Fifth and final for now short comparison with Mai, Overtly ‘girly’ in character…
YouTube had difficulties picking up ‘scan’ and rendered it as ‘can’.
What have I discovered?
Mai 2 can tend to lisp - the last ‘t’ on a lot of word is replaced with a ‘d’ - maybe this is something the American’s want but not the English ! When ‘t’ is at the start of a word it tends towards ‘th’ and makes the vocal affected with a speech impediment. The ‘ay’ sound in raise was rendered as ‘e’ [bed] and the word sounded more like ‘rez’.
If you manipulate a vowel timing you can reset it by double clicking the line between vowels/consonants. If you want to reset the volume just double click the area that changed colour.
The mouth opening strip shows a faint line where the mouth opening is set by the program - you have limited ability to manipulate this so it seems and it looks far too busy to represent real life. I don’t know anyone who sings long notes with their mouths moving so much as this line indicates!
Although the rendering was indeed quicker in SV2 (I increased the rendering time in SV1 by using groups of smaller notes [ALT+G]), I found it took much longer to go through the pronunciation to get a resemblance of what I wanted. The voice needs setting correctly to sound good from the get-go!
I’m shelving Mai for the moment (after all ‘she’ is a Japanese voice - perhaps I’m expecting too much???), and moving onto Kevin…
As before, I allow YouTube to guess the lyrics - no problems with Kevin SV1, but Kevin SV2 had ‘selection’ interpreted as ‘section’. Much better than Mai I feel.
Second test with Kevin 1 vs Kevin 2
YouTube made a mistake with SV2 interpreting ‘crutch’ as ‘crush’… But much better than Mai…