SynthV Synthesizer V Discussion Thread

[Leon]

I've already heard the usual complaints that they sound too samey (mostly Cong Zheng sounds like Qing Su and/or Weina, and Xuan Yu like An Xiao) but I have to say I like them both! Cong Zheng more in this case. Maybe it's an unwelcome opinion, but I was always very lukewarm on Qing Su. I like Cong Zheng's tone much better! Xuan Yu is also nice, even if I'm not quite as in love. He has some nice power in some spots in that demo.

 

[Maxi]

I really like Cong Zheng tone ! aah

 

[Granata]

Oh my god I love how Cong Zheng sounds so far, she might be one of if not my fav letter depending on how the final product will be like

 

[Blue Of Mind]

I love Cong Zheng's tone, but at the same time I can see why some people think nearly all of Dreamtonics' VBs sound similar to each other. I don't think it helps that the letter VBs come out with regular frequency (unlike, say, a smaller third-party dev like Eclipsed Sounds right now, who are experiencing hype like no other with Asterian).

 

[DefiantKitsune]

I also don't think that it helps that they have LITERALLY NO distinguishing design characteristics. Like, I don't think all VBs need to have an avatar or whatever, but having several paid VBs with literally just letters makes it pretty hard for any to stand out at all.

 

[peaches2217]

I also don't think that it helps that they have LITERALLY NO distinguishing design characteristics. Like, I don't think all VBs need to have an avatar or whatever, but having several paid VBs with literally just letters makes it pretty hard for any to stand out at all.

That's my main issue. They have a couple of VBs I'd consider standouts, but considering they're all represented by visuals that are difficult to distinguish from one another given how many of them there are, it's hard to remember which is which. You get to a point where you can't bring yourself to care or really be invested anymore.

 

[Tortoiseshel]

I also don't think that it helps that they have LITERALLY NO distinguishing design characteristics. Like, I don't think all VBs need to have an avatar or whatever, but having several paid VBs with literally just letters makes it pretty hard for any to stand out at all.

And they seem to be running out of color schemes, too. Xuan Yu's logo has almost the same colors as Mo Chen's. At least VY1 and VY2 had items associated with them... Actually that just gave me an idea! Honestly, I didn't even register these new guys as being rock vocalists until I read it in the descriptions. Maybe instead of a basically meaningless two-letter logo that indicates literally nothing, they could have them represented by one or more items associated with the genre they're intended for, or even just the general vibe they're going for?

If Feng Yi's intended to invoke a Beijing Opera vocalist, give her a prop associated with that like a paper fan, or one of those fancy headdresses, or those flags they wave around. Give a rock vocalist like... a cool skull or something, just anything to keep them from getting totally lost among Dreamtonics' all-you-can-eat alphabet soup.

 

[___]

I didn't even register these new guys as being rock vocalists until I read it in the descriptions. Maybe instead of a basically meaningless two-letter logo that indicates literally nothing, they could have them represented by one or more items associated with the genre they're intended for, or even just the general vibe they're going for

Or not demo them on songs that can barely pass for rock songs

but don't get me wrong, I agree with you, honestly the graphic designs need big rebranding imo, they're ugly to look at and are limiting.

 

[lIlI]

SynthV Pro is currently topping DLSite's PC software chart, could this be related to Mai's impact? I also saw a Japanese producer go semi-viral on Twitter while testing her voicebank, based on the replies, Japanese users seemed surprised by her level of realism.

 

[pico]

could this be related to Mai's impact?

I think so. A bundle-in VB is a big deal.

Whether or not this continues, we’ll have to wait and see.

 

[MillyAqualine]

And they seem to be running out of color schemes, too. Xuan Yu's logo has almost the same colors as Mo Chen's. At least VY1 and VY2 had items associated with them... Actually that just gave me an idea! Honestly, I didn't even register these new guys as being rock vocalists until I read it in the descriptions. Maybe instead of a basically meaningless two-letter logo that indicates literally nothing, they could have them represented by one or more items associated with the genre they're intended for, or even just the general vibe they're going for?

If Feng Yi's intended to invoke a Beijing Opera vocalist, give her a prop associated with that like a paper fan, or one of those fancy headdresses, or those flags they wave around. Give a rock vocalist like... a cool skull or something, just anything to keep them from getting totally lost among Dreamtonics' all-you-can-eat alphabet soup.

I'd even say they even switched Saki's logo being outstanding with that tartan-styled colour scheme to a magenta colour (while I love this kind of colours, I fear it kinda makes her mash and blend too much into the other letters despite being the one that started it all)

 

[nightmare]

Not only Dreamtonics' voicebanks don't have recognisable images, but their demos are almost always bland, and make the voices sound boring. I don't think girl/boy-next-door voices were ever that popular in the vocal synth community, and DT never shows off the cool things their voice databases are capable of. Xuan Yu and Cong Zheng both have nice voices, but I don't find anything interesting or special about them.

 

[pico]

The way SynthV works makes me feel like every voicebank basically "sings the same way" if you allow the model to 'drive'. If you turn down the influence of the AI, you regain some of the flexibility of the concatenative synths, but you do lose another element of the realism of the SynthV models.

On the other hand, if you let the AI do what it does best, you'll end up with an extremely convincing render, but the... quality? The... motif? I don't really know what vocabulary to use to describe what I'm trying to say, but hopefully it'll make sense, of the voice, will be the same throughout any voicebank you plug in there.

Yeah, it'll be significantly different sounding if you put a seriously different VB in there, like a bass voice (ASTERIAN) or a whisper voice (Chifuyu) or something of a different quality altogether (YUMA) but if they were actors in a play, I think the feeling is as if the singers are delivering their lines basically the same way, even in these edge cases. (Someone mentioned on the boards that they felt SOLARIA's release version felt like a disney princess-ified version of her voice provider/unfinished VB demo songs, and I think at least part of that effect can absolutely be owed to the current state of SynthV's model rendering. There really was, literally, less data for it to "drive" on back then, lol. Doesn't every singer kind of have a disney princess-type approach to singing? Even Feng Yi, who has an "opera" vocal mode?)

This is definitely SynthV working as intended. SynthV is hard-focusing a certain market that I think they feel does not always have overlapping sensibilities with the existing vocal synth doujin music contingent. So for these demos, I think partially at fault is the realism itself. There might be a different VB in the cartridge slot, but the model is singing each song the same way. And it's a model that they've definitely trained to sing the same kinds of songs they always demo realistically. They want to present these products under the most favorable conditions, so this totally makes sense.

But if Dreamtonics wants to express to the user how they can imbue their own emotion and artistic sensibilities into the words their voicebanks sing, it's definitely the wrong angle for all of their advertising, right?

The incredible quality of the AI has a 'wow' factor that can't be understated, but as an instrument, I think it'd be cool for Dreamtonics to go further in depth as to how SynthV works as an instrument instead of just as a simple plug-and-play AI model. But maybe that isn't where the money is right now?

I think whether or not they follow up on that will be a big factor as to whether or not SynthV serious long-term staying power among DTM people. (And I really do think it can but we're still in the beginning days in the big scheme of things)

 

[nightmare]

Yeah, it'll be significantly different sounding if you put a seriously different VB in there, like a bass voice (ASTERIAN) or a whisper voice (Chifuyu) or something of a different quality altogether (YUMA) but if they were actors in a play, I think the feeling is as if the singers are delivering their lines basically the same way, even in these edge cases. (Someone mentioned on the boards that they felt SOLARIA's release version felt like a disney princess-ified version of her voice provider/unfinished VB demo songs, and I think at least part of that effect can absolutely be owed to the current state of SynthV's model rendering. There really was, literally, less data for it to "drive" on back then, lol. Doesn't every singer kind of have a disney princess-type approach to singing? Even Feng Yi, who has an "opera" vocal mode?)

This is definitely SynthV working as intended. SynthV is hard-focusing a certain market that I think they feel does not always have overlapping sensibilities with the existing vocal synth doujin music contingent. So for these demos, I think partially at fault is the realism itself. There might be a different VB in the cartridge slot, but the model is singing each song the same way. And it's a model that they've definitely trained to sing the same kinds of songs they always demo realistically. They want to present these products under the most favorable conditions, so this totally makes sense.

Every company has its own sound and style. Rikka, Karin and Chifuyu have a similar sound, and so does DT's alphabet soup. There is definitely an overarching quality to SynthesizerV voicebanks that's due to the model it uses, but that's not what makes them sound boring. It's how the singers are directed at each company that matters most.

Emma Rowley was trained to sing in the style of Broadway's musical theatre. She was likely told to use that style when recording Solaria, instead of the pop style she sings in often. Hence the "Disney princess" style of singing - it's just Western musical theatre that hasn't been translated into synth yet. With Asterian, we can see that Eclipsed Sounds' focus is on filling in niches.

Rikka, Karin and Chifuyu all have a similar nasal sound due to how they're voice acted. Meanwhile, Dreamtonics' voice directors seem to ask their voice providers to sing in the most neutral way. But no matter what style a voicebank is recorded in, Dreamtonics should stop being scared to show off what their voicebanks are capable of. Natalie's demo is just criminal, for example.

 

[___]

Every company has its own sound and style. Rikka, Karin and Chifuyu have a similar sound, and so does DT's alphabet soup. There is definitely an overarching quality to SynthesizerV voicebanks that's due to the model it uses, but that's not what makes them sound boring. It's how the singers are directed at each company that matters most.

Emma Rowley was trained to sing in the style of Broadway's musical theatre. She was likely told to use that style when recording Solaria, instead of the pop style she sings in often. Hence the "Disney princess" style of singing - it's just Western musical theatre that hasn't been translated into synth yet. With Asterian, we can see that Eclipsed Sounds' focus is on filling in niches.

Rikka, Karin and Chifuyu all have a similar nasal sound due to how they're voice acted. Meanwhile, Dreamtonics' voice directors seem to ask their voice providers to sing in the most neutral way. But no matter what style a voicebank is recorded in, Dreamtonics should stop being scared to show off what their voicebanks are capable of. Natalie's demo is just criminal, for example.

Direction plays a big role in how voicebanks turn out, I totally agree! But I hear Pico in that, there's something inherent to the model that makes the singing style samey.

I think they were speaking less about color of the voice, but the capturing of the singing style, which SVS simply doesn't do, and it becomes apparent when you contrast it next to something like Cevio AI.

 

[pico]

I think they were speaking less about color of the voice, but the capturing of the singing style, which SVS simply doesn't do, and it becomes apparent when you contrast it next to something like Cevio AI.

You did a better job explaining it than me. Thanks :)

And for what it’s worth, I do very much agree that the direction of the voices has a big impact on the final product (I really like TOKYO6’s characters and Eclipsed Sounds’ niches) with this caveat in mind.

 

[Blue Of Mind]

It's probably worth mentioning that Vocaloid's engine also renders its VBs in a way that gives them a distinct sound, to the point that some people have complained about VBs sounding too "Vocaloid-y" or obviously artificial in the past. It's hard to explain as well, but the simplest way to put it is that while SynthV focuses more on realism, to the point that even the voice acted VBs sound more realistic than expected, Vocaloid still has a characteristic sound that can be heard even on more realistic VBs. In short, their engines are polar opposites in what results they're trying to achieve.

SynthV guns for realism first and foremost, so unless you intentionally try to create a unique bank (like Asterian), chances are it might sound similar to another bank that already exists. This is probably why a lot of the letter banks sound similar to each other (aside from Dreamtonics not pushing the envelope internally when it comes to voice types.)

 

[Tortoiseshel]

Or not demo them on songs that can barely pass for rock songs

You know... I didn't want to say it, I thought maybe I was just biased towards the heavier stuff and saying something like "This isn't real rock!" would just make me sound like a big jerk. So I'm glad other people thought it too!

 

[TheStarPalace]

SynthV's realism is good at impressing people on YouTube who haven't heard vocal synths in years and has gained notoriety in the Japanese, Chinese, AND English markets (not easy!) But it genuinely feels like they want to present their first party voices as neutrally and blandly as possible so other users can fill it in with their own ideas. Not bad, but I and the rest of this thread got bored of the alphabet at least a few letters ago Ryo and Kevin releasing on the same day already felt like a lot! I've wondered for a long time if they keep reusing the same training data to make these AIs with how samey the direction is. Cross lang seems to have improved but it sucks for people like me who bought Ryo and not one of the newer letters so I was stuck with editing each of his phonemes individually

 

[morrysillusion]

one of the things that really stands out to me as we get more letter voices and their samey-ness is the fact that maybe it wouldnt be so obvious if there were... more third party voices??? i know we have some in development, but in terms of overall releases of AI voices (which at this point is the expectation, for them to be AI) there really hasnt been much! its either been long awaited vocal mode updates (some of which have been dissapointing) or announcements for voices that we are just waiting to release. sure i wouldnt say we have loads coming for engines liek Cevio, but i guess the issue to me is i feel like the fast rate in which we get dreamtonics voices makes this sooooo obvious, and makes the issue w their neutral angle so much more blatantly obvious. if we have a decent variety of voices coming from other companies that are more likely do something interesting, we'd at least be able to say we had some options- but we dont really.