#vocalochanger

[liure00]

【Fuiro】Rolling Girl -Rock Arrange- (Short Ver.)【Vocalochanger Cover】

[krathia7]

Synth V | VOCALOID?

Many people keep saying how outdated is VOCALOID6. Is that true? Well, that's purely in the eyes of the beholder.

If we think about realism out of the blue, without too much effort other than polishing an already good outcome... Yes, Synth V is way better. And, don't get me wrong, being superb at is also takes time.

Then we have VOCALOID6. In essence AI voices here are a.... polished version of how VOCALOID5 was, but without the voices going out of tune while "trying to add a human way of singing". However, you have very little control, pitch, dyn and something close to tension... but in barebones. Great things can be done, but is less comfy. HOWEVER VOCALOID6 also has a new feature: Vocalo Changer. It seems useless unless you have plenty of dry acapellas, you don't know how to sing or if you don't have a proper environment to record. But it can capture little details, like MJ ones, with the vocaloid voice.

I wish YAMAHA were adding more features tho. I would kill to have an option to convert vocalo changer outputs to VPRs, because it's actually really accurate to capture pitch. Or I will love to have more parameters to play with.

[balzabul]

Original song by Norihiko Hibino for Metal Gear Solid 3; svp file by scenikeight. Reference vocals are Eleanor Forte.

Trying to use the V6 default gang more but unfortunately they mostly sound best through vocalochanger lol

[lesbian-forte]

Criticisms of Vocaloid and why I like SynthV

I'm not trying to change anyone's mind here, but I would like to say my piece after certain takes seem to miss the point entirely. This might be a bit of a rant.

Vocaloid has gone stagnant in recent years. Yamaha doesn't care. Yamaha doesn't need Vocaloid and is a large corporation that gets much more money off of their DAW software and actual instruments as opposed to something as niche as vocal synths that are both only big in Japan and also only if they're in the top ten or so.

Yamaha stopped putting effort into Vocaloid during the V4-V5 transition. There is a reason V4 has so many cancelled voicebanks. Several developers were working on V4 and Yamaha rendered their devkit suddenly worthless. Devs would have to purchase a V5 devkit and start work over, or quit Vocaloid. And as vocal synth companies are generally very small, few of them would want to continue or even be able to afford it.

So they moved. Miku splitting off for Piapro gave them an opening, and others started looking for alternatives. Then IA went to CeVIO. And more and more. And by the time V5's sun was setting, all the third parties that worked on that were gone too.

But for a while, you didn't hear much from most of them. If a company released a V4 at the tailend of its lifespan or a V5, they had to wait for exclusivity to expire. And Yamaha's exclusivity deals are harsh (ending distribution of existing song voicebanks in the case of utaus with the same VP) and long, borderline predatory. So voices that companies wanted to update couldn't receive them until those expired, or else refresh that deal and stay constrained by a company that didn't even want to bother with them.

So, come V6, Yamaha was desperate. Internet Co had made an ultimatum that if a Vocaloid 6 didn't come out soon, then they'd be going too. That was their last, and after Crypton packed their bags, most important third party. So they accelerated their plans and looked at what the new guys were doing to be so successful.

They took the wrong lesson.

AI is not inherently better. Sample-based voicebanks will always have their place. Traditional samples can allow an unnaturally large range and harsher voice acting than would be possible to maintain. AI is more accurate to the voice provider, and you have a greater degree of freedom with its tone, plus updates and additional features are so much easier- but Yamaha took 'AI' at face value and made a low-quality copy that sounds significantly worse than prior Vocaloid versions and pushed it with Gumi. They could have stuck to improving their concatenative synthesis render quality further- that's what SynthV started as, R1 was just a very well-rendered sample-based program that is probably just a fancy utau under the hood.

But Vocaloid jumped on the bandwagon by doing the absolute bare minimum and claiming the ear-grating engine noise that can cause actual nausea is remaining faithful to the 'Vocaloid sound' even though styrofoam on the mic and a sometimes pleasant metallic twang sound nothing alike. They didn't improve accessibility, V6 has the same stability issues as V5, and the shiny new feature Vocalochanger is just RVC but worse.

Then, less than a year after product launch, they start up VxB and don't do anything to improve the software they're actively selling. Internet Co themselves called this out in the form of a Gumi tweet. Then Internet Co got in talks with Tokyo6 and saw a possible out, so they gave it a go. They're still under contract by Yamaha so what they can do is limited, but we saw them stray as well. And the result is a much better quality version (though arguably still worse than her V4) despite being an exact port.

We're still getting a Gumi Solid V6 because V6 can't do emotions and they still have to be separate banks, and VxB still got a major update even though it's dying in April with radio silence for V6 development, while CeVIO/VoiSona is releasing 2.0s that get major acclaim like Ci Flower's reputation totally getting turned around, and SynthV is sitting pretty with several voicebanks announced and several coming out in December alone, and the most recent in-progress update including both voice-to-midi (which is what vocalochanger should've been) and Spanish.

I do not like V6, V5, or Yamaha. It could've been amazing for Gumi or Una to get updates so they'd have crosslang (or better crosslang) capabilities, as I work in English. But the result was extremely poorly implemented and Yamaha has made no effort to fix that.

I use SynthV all the time, I'd do the same for CeVIO if it offered crosslang as well rather than just dictionaries and a couple English banks. I'm not against trying new things. But I either want the other programs to have the things to suit my needs in a quality manner that's intuitive to use or for the voicebanks I love to get versions on programs that already do. It's not that complicated.

The jokes and the yammering from rabid SynthV fans dissing Vocaloid can get to be too much sometimes. But you have to consider where that actually comes from. It's in response to suddenly being spoiled with a cheap, accessible, high-quality program when the expensive, poorly constructed, difficult one has been dominating the market with anti-consumer, anti-third party practices for years.

P.S.: Also you can do robotic tuning and mixing on realistic vocal synths, it's called doing the same thing as before and then adding it in post. You think utaites swallow vocoders or something? No, they just use different tools to get the same result as engine noise. Not fighting the voice when you're trying to go for realistic vs manual tuning and adding some very easy effects on when doing it the other way around is better, actually.

[cindersnowsstuff]

VOCALOID6 Free Trial Review

The short version: I didn't like it.

The long version: I tried out the VOCALOID6 free trial over the weekend and instantly hated it. Not being able to draw the pitch line with the pencil tool is so disappointing, and making pitch edits feels sluggish. The default AI tuning doesn't really do much either, and it makes the vocals sound drunk. It's crazy how lame this feels when it's so easy to make something sound good in SynthV. I think the worst part is that AI voices don't have the voice color options (like growl and character) that non-AI VOCALOIDs do. I double-checked the manual just to make sure I wasn't blind. Shouldn't AI voicebanks give you more options instead of less? I should note that I sold my past VOCALOID stuff, so I wasn't able to try any non-AI voices in this editor.

V6 is fine if you want a robotic/autotuned vocal sound, but past editors can already do that. The Vocalochanger is certainly there. It exists. I'm not sure what the point is though. If you can already sing well enough use the voice changer and not have it sound like ass, then why even use a vocal synth in the first place? Maybe I'm missing something. I was surprised to see that the custom style presets I made in V5 were saved to V6 when I had uninstalled V5 previously. That's something nice I guess.

V6 is very underwhelming, and I don't see myself buying it unless there's some insane 90% off sale or something (and they also make Lily or Mayu AI so they can sing in English lol). It's not really an improvement over V5. I've never used V4 or earlier, but I've heard others say they prefer those over V5 and V6. I've only started using vocal synths in the past year, so I'm definitely not as knowledgeble as others. This is just my experience. V6 probably is fine once you get used to it, but it doesn't seem worth it to buy or upgrade at this point. Maybe someday they'll improve things like pitch editing and voice color options, but I'm not getting my hopes up. SynthV is the only other AI editor I've used, and V6 doesn't even come close to matching it.

[red-dyed-sarumane]

i know im the one constantly complaining about people who say sv is the superior software & every vocal should have an sv bank but im a voca scene defender first & foremost & if u start trying to exclude it and other "ai" programs just because u let the internet train u into recoiling from any mention of ai then maybe u shouldnt be starting events u claim are centered around the community as a whole. its ai in the sense of an algorithm so the default tuning isnt quite as "robotic" it's still ur same ethically sampled voice it still requires quite a bit of user effort. it is so fucking far from one click and done auto generation. if ur saying this i know uve never worked with or seen anyone actually working with them because damn the "ai" is so off and so bad sometimes looking at default pitch lines is one of my favorite pastimes because of it. perhaps its making the process just a bit easier but its still so immensely different from going to an ai generation site to make say squidward sing a britney spears song. even tools like voice to midi or vocalochanger arent really touching that. its not the same thing. if u really know what ur doing u can do everything those 2 tools do only manually its really just helping streamline the process. its not even mandatory u can forego them entirely. are you really going to sit there and say songs with hours of effort put into the vocal synth, to get the notes in, tune everything, keep it timed, edit the phonemes so it sounds more like the intended words, all of that means nothing because it has a few tools with a name you dont like. im not sitting here and arguing for ai learning like chatgpt or whatever. there is in fact a difference from mass sourcing information or art from nonconsenting and often completely unaware people and taking a few more hours of samples from a paid consenting voice provider to make an instrument sound a little more like them with a little less user effort. these are still SYNTHS these are INSTRUMENTS just because one guitar plugin sounds more like a real guitar than a different one doesnt mean u need to boycott the more realistic one. also, hell, for a lot of if not all of (i havent kept up with all of them) v6 voicebanks there are, in terms of use, things that they tell you not to have them sing about. very specifically saying thats what they do and do not consent to having their voices used for. and it Does vary per synth its not a cover all. some of u need to learn to stop jumping at buzz words and actually learn what it is ur dealing with.

[kurgy]

at the risk of sounding insane I want yall to know as I'm learning this that Cypton a Yamaha company responsible for vocaloid recently announced the debut of a new voicebank on their v6 engine, a previous UTAU and first official commercial furry voicebank on the market. in the demo for his voice, vocal synth fans discovered that the voice was actually synthv Yuma through the vocalochanger. the dev team then admitted to this to some degree? and released a new demo. fans quickly pointed out this also was not a vocaloid, but in fact an RVC generative voice ai. the furry character itself is deeply hated by the community, as is the creator, and no one knows why this voicebank was chosen for commercial release above many others, but even more confusing is that the voicebank seems to not exist

[auspicious-voice]

Voicebank Development Log (11/24/2022)

Oh hi! It's been almost SIX months since I last written a post related to voicebank development logs. I did say I was taking a break from making voicebanks for a while, but here I am slowly working on new voicebanks for my UTAUs...

UTAU Voicebank Development

It's been a hot couple of months since Maria and Mario's latest Japanese voicebanks were released for download, and they were met with positive reception so far! I'm happy that the voicebank release was very successful :D

That being said, I am slowly working on new voicebanks for the two, but development will be in full swing starting mid-December. I am excited about these new voicebanks because Maria and Mario will be receiving English voicebanks for the first time!

These voicebanks are in Arpasing format and are expected to be released by next year. The oto.ini configs are being refined at the moment, and I will provide more updates in the future (and hopefully some new demos)! Maria's Arpasing voicebank will be named CELADON, while Mario's Arpasing voicebank will be named TACHYON.

Hopefully when I have enough free time, I can either start recording their Korean or Chinese voicebanks...

DiffSVC Model Development

DiffSVC is the free Vocalochanger from what I've seen so far, and while I didn't really want to make AI-related stuff for my UTAUs, I ended up doing it anyway.

The DiffSVC models that I am working on for Maria and Mario are currently in the works, with their datasets consisting of their UTAU recordings. Each dataset is approximately one hour and thirty minutes long, so that's a lot of data to feed to the training notebook.

Maria's DiffSVC model is being developed before I work on Mario, and is using the nyaru pre-trained model to train.On previous tests so far, she's sounding pretty good, but I added new recording data to her dataset, and with the new DiffSVC updates, I had to restart my progress. The main goal is to get to 200k steps to ensure a properly trained model that sounds good on output, and she's at about 173K steps in total! :D

I'm not sure if I will release Maria and Mario's DiffSVC models in the future, and I have no idea what I'll be naming these models as. That aside, it's about time Maria and Mario receive AI voicebanks, so their DiffSVC models will be the first step in this direction! (I'm also looking at DiffSinger and hot damn this singing AI shenanigans is making me really interested)

[tomedonaki-hakujou]

ZETHIEL ZERO AI for Diff-SVC

ZETHIEL ZERO, or ZETHIEL_LV0 is a representation of my voice before HRT, and the younger version of Alchemist ZETHIEL. He is a young teenage boy with a strong, energetic voice -- but a wavering heart.

Using Diff-SVC, this technology allows me to preserve and revive my voice from before HRT.

[video text transcription: ZETHIEL ZERO AI, in development, Diff-SVC voice model, mezzo-soprano male voice, 40 minutes, 25k steps demo, English/Japanese voice data, Queen - Kanaria]

more about Diff-SVC and my process under the cut:

Diff-SVC is essentially a voice morphing AI rather than a singing synthesizer AI (think VOCALOID 6 VocaloChanger or TikTok voice filter; as opposed to NNSVS and Synthesizer V.)

What makes Diff-SVC have an easier level of entry as opposed to NNSVS/ENUNU is that much of the work is automatic than hands-on. All you need to do is provide voice data (singing or talking works!) and the correct file directories, and the AI will learn the voice input to be able to replicate it. Then, you can use voice models to sing whatever singing (or talking) file input you want. More info here.

Steps correlate to the amount of time the AI has spent training on the voice. Theoretically, the more steps, the more realistic the input. I plan to release ZETHIEL ZERO at around 100k steps, with options to use lighter versions like 25k.

For ZETHIEL ZERO, the singing input is all singing data from before I went on testosterone. At that time, I was fighting so much to assert my maleness despite my voice. Likewise, now, I will assert ZETHIEL ZERO as a male vocalist to break the expectations of what a "male vocalist" should sound like.

[witch-renna]

Vocaloid 6 VocaloChanger test.

SVP by Hugwalk (I tuned it in Synth V bc I hate V6 LOL)

[azover]

Things like vocaloid (including vocaloid ai and vocalochanger), synth V and utau voicebanks: cool and good, they're just tools/musical instruments

Shit like making ai voices out of real people, esp politicians and people in power, or out of someone you know solely for the purpose of making them say what you want them to say: what is wrong with you

Why are people so lax about AI voices, it's honestly way scarier to me than AI art

[krathia7]

【SARAH】All I Want for Christmas Is You【VOCALOID 6 Cover】

[balzabul]

original song by braden ross ft. jay kim

its terrible glitchy effect happens to kind of work for this song anyway but man vocalochanger is just. not very good. at all. thanks anyway yamaha

[samurailogic]

I hate RVC and vocalochanger covers lol

[aseriesofunfortunatejan]

I'm sorry but I feel like everyone should be freaking out about how I can sing! And have the tuning of MY SINGING! Be translated to SynthV!

(don't. judge my singing i guess)

We keep getting great news, like SAROS' release and Gumi SynthV, but this is genuinely a game changer. And if you want to compare VOCALOID6 to SynthV, I think this new function means at least as much as Gumi's port. I'm already seeing people state they like this more than VOCALOchanger because the latter sounds too much like the original singer and too little like the selected voicebank... which honestly... is true. This above sounds like Mai, not myself.

Edit: I shouldn't have only tested Mai... I've always said I felt my singing was somewhat similar to Qing Su's and I'm so happy to hear her singing like ME! And then I can GENDERBEND MYSELF

[fearofevrything]

boo!

it's almost halloween and while it's not exactly a spooky song, it's close enough.

weird matchup here tbh but I wanted to test vocalochange so i popped the tuned chifuyu version into v6 and switched the high parts to una and...well it ended up sounding better than chifuyu alone

thank you for watching! <3