Maybe this is because I recently purchased V5 a few days ago, and use the preset styles, but in terms of rendering and sound quality, I do believe V5 sounds better than all previous editors combined. Vocal synths really are what you make of them. Over a few years, I have heard varying degrees of quality from many voice synthesizers. So there are some people who can create amazing vocal tracks with a voice synth, and others that barely do anything to the vocal. But I do find that it's challenging to edit V5's parameters smoothly without holding down the shift key. It takes a while to generate realistic results results, and albeit there are still some pronunciation issues (as well as issues with sounding realistic), I still believe VOCALOID is one of the better vocal synths out there. But there are other alternatives as well, though I don't know as much about those since I have only used VOCALOID4, Piapro Studio for V4x (which is powered by VOCALOID4 API), and VOCALOID5.
Also, if anyone is interested, I have linked some realistic original songs by Nishin, various VOCALOID5 compatible vocals. There are also some Japanese vocal usages I know of that sound nice (using realistic, non-voice acted vocals), but I will not post them unless someone asks about them.
Epilepsy warning for this first one, flashing images. This is a VOCALOID3 voice used in V5:
https://www.youtube.com/watch?v=P0D768PAmHE
Here's another song:
https://www.youtube.com/watch?v=GyqOANpkWxk
Fire your singers folks, Vocaloid 5 is here!
-
- KVRAF
- 3231 posts since 18 May, 2003 from Sweden
Interesting examples, that simultaneously reveal both the pros and cons of the current state-of-the-art of synthesized voices.
I quite liked the first song, especially. But I wouldn't have listened more than maybe 30 seconds without the 'subtitles', since 'her' diction, or rather, lack of it, often makes it very difficult to understand what 'she' is singing. In fact, the animated titles add their own weight to the lyrics and 'her' slurred diction in this context becomes some sort of feature instead of being a 'bug'.
The videos seems to me to a very valid way to use currently available digital singers in creative way. The lyrics had better be good, though…
I quite liked the first song, especially. But I wouldn't have listened more than maybe 30 seconds without the 'subtitles', since 'her' diction, or rather, lack of it, often makes it very difficult to understand what 'she' is singing. In fact, the animated titles add their own weight to the lyrics and 'her' slurred diction in this context becomes some sort of feature instead of being a 'bug'.
The videos seems to me to a very valid way to use currently available digital singers in creative way. The lyrics had better be good, though…
If it were easy, anybody could do it!
-
- KVRist
- 395 posts since 6 May, 2020
These examples are indeed wonderful. And as the comment above, I think it really highlights what would need to be addressed when synthesizing vocals, at least when it comes to the English language or strong vocal performances. The best examples, as well as the most popular adaptation of such software, would be the Vocaloid culture utilising the Japanese language as the outlet. I myself am not that familiar with it extensively, but the examples I've seen showed close to realistic performances. Why that might be the case, at least in my opinion, boils down to genre differences, as the more pop-inspired high pitch vocals don't suffer from artefacts as much, or at least it seems to be the case.
Really interested to see a day, when fully computer-generated vocals can compete with human performance when it comes to emotion. That might be far off, but seeing technology improve is fascinating.
Really interested to see a day, when fully computer-generated vocals can compete with human performance when it comes to emotion. That might be far off, but seeing technology improve is fascinating.
Take care 
-
- KVRian
- 1320 posts since 26 Aug, 2019
I'm curious if anyone here has experimented with using a formant filter effect on a vocaloid track?
Seems to me that a formant filter applied to certain vowel combinations could smooth over some of the most grating synthetic aspects I hear in some of the voices I've tested. In my limited experimentation, and setting aside the problems with consonants, what I have noticed with certain vowel sounds is that the most synthetic qualities of the voice are exaggerated. It varies by voicebank and is accentuated more with certain voices. I'm convinced that no amount of tweaking the phonetics will improve things beyond a certain point, and the problem seems to be especially notable on vowel dipthongs.
So the thought occurred to me that DAW automation along with a formant filter over a vocaloid track could improve the sound of vowels in interesting musical ways. I plan to experiment more but curious if anyone with more experience has gone down this road already?
The other thing I've considered is going lofi with the vocaloid voice and degrading the sound as if it were recorded from an old radio or over a telephone, which from what I have listened to is another way of approaching the problem. I may have to use both techniques. I have played around with saturation plugins like Voxengo Warmifier, and those definitely have a role here but a very subtle one in combination with other effects.
I'm not trying to cross the uncanny valley towards a voice that is indistinguishable from a human voice. I have a project that calls for synthetic voices, and there are aspects of them that I find annoying (for lack of a better term) and so I'm thinking about ways to improve the musicality of the voices to my ear. The same way I find autotune annoying. Sure you can bury the voice in a mix or do things with a vocaloid that a human just couldn't do naturally, which seems to be where vocaloids perform best, but I'm not trying to accomplish either of those. Just trying to get to a vocaloid singing voice that doesn't drive me up the wall the longer I work with it.
Seems to me that a formant filter applied to certain vowel combinations could smooth over some of the most grating synthetic aspects I hear in some of the voices I've tested. In my limited experimentation, and setting aside the problems with consonants, what I have noticed with certain vowel sounds is that the most synthetic qualities of the voice are exaggerated. It varies by voicebank and is accentuated more with certain voices. I'm convinced that no amount of tweaking the phonetics will improve things beyond a certain point, and the problem seems to be especially notable on vowel dipthongs.
So the thought occurred to me that DAW automation along with a formant filter over a vocaloid track could improve the sound of vowels in interesting musical ways. I plan to experiment more but curious if anyone with more experience has gone down this road already?
The other thing I've considered is going lofi with the vocaloid voice and degrading the sound as if it were recorded from an old radio or over a telephone, which from what I have listened to is another way of approaching the problem. I may have to use both techniques. I have played around with saturation plugins like Voxengo Warmifier, and those definitely have a role here but a very subtle one in combination with other effects.
I'm not trying to cross the uncanny valley towards a voice that is indistinguishable from a human voice. I have a project that calls for synthetic voices, and there are aspects of them that I find annoying (for lack of a better term) and so I'm thinking about ways to improve the musicality of the voices to my ear. The same way I find autotune annoying. Sure you can bury the voice in a mix or do things with a vocaloid that a human just couldn't do naturally, which seems to be where vocaloids perform best, but I'm not trying to accomplish either of those. Just trying to get to a vocaloid singing voice that doesn't drive me up the wall the longer I work with it.
