A.I.VOICE (AITalk5) General News

[uncreepy]

I have a thread on Twitter investigating the installer, interface, and voice bank trials for A.I.VOICE. I'm just going to link to the part where I uploaded examples of normal Yuzuru + his 3 emotions, then dark Yuzuru, then me using voice fusion to give his dark bank all 3 emotions.


You can check out this thread for some mildly interesting info regarding the software if you want. I will upload other fusion combinations throughout the next few days (there are a lot of combinations)!

 

[uncreepy]

Here is the Aoi, Aoi bud, and Aoi bud with emotions version:

I have a lot of other combinations left, but I don't want to record anymore today.

 

[uncreepy]

Today's tests are Yuzuru + Aoi (different combinations of them, including dark and bud). Fusion uses Aoi's voice as a pitch reference, so I had to lower the pitch parameter to 0.5 in order to make him sound masculine again (without it, he sounds very high pitched and feminine). I will probably upload a recording of this later without pitch edited because it's informative, but also sounds cursed.

 

[uncreepy]

An article from Dec 22nd says that AI Inc is collaborating with Acapela Group, "Europe's largest speech synthesis technology company" in order to expand their lineup of diverse voices in multiple languages. AI Inc is obtaining the right to sell the products made by Acapela in 34+ languages in Japan using their AITalk 5 engine (the one with deep learning). They don't mention A.I.VOICE in this article, but I feel like we can assume the English and Chinese voices coming in December will sound like or maybe even be some of the voices you can demo on Acapela's website.

You have to select a voice, checkmark the box that you accept the terms of service, press the "Synthesize" button, wait a few moments and then a red play button appears on the right side of the Synthesize button. Please test out the WillBadGuy voice, haha. WillLittleCreature is like a bootleg Yoda. There's only one Mandarin voice called Lulu. There are several flavors of English accents, too.

An article from Jan 7th says that AI Inc got a lot of worried inquiries about if the creation of A.I.VOICE means they won't do VOICEROID anymore. They say they will continue to provide technological support for AHS depending on AHS' plans, but they ask that people concerned about VOICEROID please contact AHS to find out about it instead of them.
(I really think this sounds like AHS is the ones switching to CeVIO AI and not like AI Inc cut them off or something. We'll probably only get a new VOICEROID if AHS asks, but I doubt they will anymore.)

This is the same demonstration from yesterday, except Yuzuru is in the default pitch referenced from Aoi. It sounds very bad and high-pitched, so I recommend setting the pitch to 0.5 like the previous post I made to make him sound more natural:

 

[KingPapillon]

So I am currently torn on deciding my first Japanese talking synth.
I am exploring a few options;

A) Kotonoha AI Voice, I get the newest tech on a new program, and if I preorder now I can get a cute acrylic stand, Kotonosync support unsure currently
B) Kotonoha Voiceroid2, I get a confirmed Kotonosync tts, and If I order from AHS, I can get it bundled with Kiritan Cevio AI which I am planning to have shipped very soon
C) Meika Gynoid Talk, I get two distinct talk vbs which are Kotonosync compatible, for two different character voices I love. I have no idea how Gynoid Talk itself fares in comparison to these other programs however
D) Wait for Tsurumaki Maki Cevio AI who I am definitely purchasing in the future anyways...

Is AI Voice a real step up from these previous editors?

 

[uncreepy]

It seems like suzumofu has a beta version of KotonoAssist (helps with tuning) that is partially compatible with A.I.VOICE


In this older tweet after the 1st A.I.VOICE live steam they were a guest at, they were thinking of maybe making KotonoSync and KotonoAssist work with A.I.VOICE if possible (which is what the 1st tweet I linked to ended up partially doing)


For KotonoSync, you can't type while using it because it might mess up the song rendering, so you can't really use your computer while it does that (if I remember right, it usually takes anywhere from 3 min ~ over half an hour depending on how long the song is), there are tutorials about how to use it in the Resource tab of VocaVerse if you haven't checked them out yet. (Just have it render while you go eat lunch or go brush your teeth at night or something, that's what I do)

There are no differences between Gynoid Talk and VOICEROID, they are actually the same software but with a different brand name. AI Inc sells its AITalk engine + interface to other companies and just gives them a different label. Even GalacoTalk and Otomachi Una TalkEX uses the AITalk engine. So when making a decision, you only need to compare VOICEROID/GynoidTalk/GalakoTalk/Otomachi Una TalkEX vs CevIO Talk vs A.I.VOICE. AITalk 5 is the engine being used by A.I.VOICE.

The only really noticeable differences between A.I.VOICE and anything that uses AITalk 4 (VOICEROID, GynoidTalk, GalacoTalk, etc) is:
The A.I.VOICE interface is black and white instead of blue and white and lacks a dark mode
A.I.VOICE can export as mp3 now instead of only a wav
Customers of AI Inc only have to record 1 voice bank now and deep learning calculates happy, sad, angry instead of having to record/pay for 4 voice banks older versions of AITalk required
A.I.VOICE has voice fusion (like Cross Synthesis with VOCALOID)

None of the TTS software using AITalk have an English interface yet (will probably come in December when they release the ENG/CH voices) and KotonoSync has no English interface, either. CeVIO to my knowledge does have an English interface but you'll have to buy their song voice to make them sing.

Hopefully this helps with making a decision. (Edit: Might help to try to check out videos of the voice banks you are considering to hear them talk and hear them using KotonoSync, you should probably check out both YouTube and nico because talk videos are more popular on nico)

 

[Rylitah]

A) Kotonoha AI Voice, I get the newest tech on a new program, and if I preorder now I can get a cute acrylic stand, Kotonosync support unsure currently
B) Kotonoha Voiceroid2, I get a confirmed Kotonosync tts, and If I order from AHS, I can get it bundled with Kiritan Cevio AI which I am planning to have shipped very soon
C) Meika Gynoid Talk, I get two distinct talk vbs which are Kotonosync compatible, for two different character voices I love. I have no idea how Gynoid Talk itself fares in comparison to these other programs however
D) Wait for Tsurumaki Maki Cevio AI who I am definitely purchasing in the future anyways...

Uncreepy's post pretty much has all the info you could ever want about them, but I'd like to offer my two cents on your options as well as like... the other person on this site who uses TTS synths.

A. Kotonosync support will most likely come. I think the jump from VOICEROID EX to VOICEROID2 is actually bigger than the jump from VOICEROID2 to A.I.VOICE, since A.I.VOICE visually, at least, looks the same and the only new real feature it has that VOICEROID doesn't (that would actually impact Kotonosync) is voice fusion. But since Kotonosync as it is allows you to utilize VOICEROID2's custom preset option and render with that, I don't think it'll be particularly difficult to update Kotonosync to include it. But that's just from an outside perspective from someone who has absolutely zero experience in actually coding or programming these things, and of course it's up to Suzumofu regardless. But they're a very big Kotonoha fan, so I wouldn't be surprised if A.I.VOICE support for Kotonosync isn't too far off in the future anyway.
B. Kotonoha V2 is more or less the same as Kotonoha A.I.VOICE aside from the new voices. It would be to your monetary benefit if you were to preorder A.I.VOICE this month since the preorder price is cheap, but if you're certain you will get Kiritan soon, Kotonoha V2 is the same price as A.I.VOICE's regular price (after tax and if you're not getting the AHS user discount - if you are, V2 will be cheaper).
C. Like Uncreepy said, Gynoid Talk and VOICEROID are the exact same. The big difference here will be quality between the two companies, and the Meikas.... are not exactly high quality (at least compared to ones from AHS; also Hime's Talk and Vocaloid voices have different tones)? You can find examples on Nico like Uncreepy said if you want to compare, or I can render something for you if you want since I have them all. If you're sure you want to start off with two vocals (Kotonohas vs Meikas) then the Meikas would give you two different ones immediately; the dialect difference between the Kotonohas most likely will not matter to you for a while as a beginner to the language.
D. Also a good option, since Maki will have both Eng and JP banks. There just probably won't be a program made to make her talk banks sing (or maybe there will be, but it won't be anytime soon), but she'll also have Synth V banks to use if you wanted to buy those as well.

I know someone is working on an English patch for Kotonosync. I'd like to not divulge any more details since it's just a side project and needs approval from Suzumofu if it were to be released, but that is a possibility in the future.

I think all of these are strong options, though we don't have any idea of when Maki will release at this time. If you're definitely getting Kiritan soon, I'd just get the Kotonohas with her to at least cut down on shipping, but in the end it's up to you and they're all good choices. Good luck deciding!

 

[___]

C) Meika Gynoid Talk, I get two distinct talk vbs which are Kotonosync compatible, for two different character voices I love. I have no idea how Gynoid Talk itself fares in comparison to these other programs however

If bigger talking synth fans could chime in I'd appreciate it but I've heard Mikoto's talk ( along with Flower talk cuz I see them mentioned in this context together often ) are like...super choppy? And I see them often compared to those really old early 2010s Voiceroids in terms of quality. But this is from outsiders perspective fghjkhj.

 

[Rylitah]

If bigger talking synth fans could chime in I'd appreciate it but I've heard Mikoto's talk ( along with Flower talk cuz I see them mentioned in this context together often ) are like...super choppy? And I see them often compared to those really old early 2010s Voiceroids in terms of quality. But this is from outsiders perspective fghjkhj.

I would put them below Maki in terms of quality, which actually doesn't sound as bad as it seems since Maki's actually really good for a product released in 2010 and can stand alongside the newer Voiceroids just fine. But there's a lot of artifacts in their voices and they sound very synthesized, Mikoto in particular. I don't own Flower Talk (yet) but from what I've heard she's even worse than Mikoto orz.

But yeah, there's no problem if you really like them, but if the characters don't matter to you, practically any Voiceroid aside from the Tsukuyomis would be higher quality.

 

[KingPapillon]

Thanks everyone for the help! All of this info was great, I might go with AI Voice Kotonohas after all, and I am still def getting Maki for Cevio AI and Synth V too.

I just got Meika V5 very recently, and the thing I've learned about them rather quickly was how weird they are in quality. They can sound super natural post mixing and tuning, but there is so much weirdness going on in their voicebanks. After listening to some talk clips, I can see that it carries over a little, except its rougher since tts synthesizing is a lot more fickle in terms of audio source quality.

 

[Rylitah]

Yukari and Akari will be on A.I.VOICE. Release date sometime this year.

Also, reminder that next week on Thursday is the livestream where Yuzuru's voice provider will be revealed.

 

[___]

Huh youd think Yukari would get new talk on Cevio considering she has song voice there.

 

[uncreepy]

Part of me is happy they're on A.I.VOICE instead of CeVIO AI because I've been really disappointed in that engine, plus you can use fusion so the results will probably be interesting. Weird decision from a marketing stand point, though.

 

[Rylitah]

Yuzuru's voice provider is Matsuura Yoshiyuki.

 

[uncreepy]

I am writing a summary for the live stream on this thread:


I'm about 20 min in right now, but I'm going to gradually update it throughout the day (I don't feel like listening to it all at once).

 

[Rylitah]

A.I.VOICE can now be purchased digitally.

 

[___]

@uncreepy it's your time

 

[uncreepy]

@uncreepy it's your time

XD I actually bought Yuzuru immediately last night as soon as they announced it.


By the way, I got clarification about upcoming release dates for Rei, Yukari, and Akari. I had been wondering if Akari and Yukari were the mysterious July releases, but I guess they are not! (w00t, more voice banks!)

 

[Mellhurst]

Does this do real-time conversions (for vtubing/acting on Discord, etc)? What about Voiceroid, CeVIO, Synthesizer V, A.I.VOICE, Crypton?
I can probably muddle through these programs being in Japanese only with ongoing learning, and translators...

Thanks!

 

[uncreepy]

Does this do real-time conversions (for vtubing/acting on Discord, etc)? What about Voiceroid, CeVIO, Synthesizer V, A.I.VOICE, Crypton?
I can probably muddle through these programs being in Japanese only with ongoing learning, and translators...

Thanks!

Voiceroid, CeVIO Talk, A.I.VOICE are text-to-speech. You have to type the text in Japanese, there are no English voice banks yet (Tsurumaki Maki is getting an English TTS voice soon, but she has a Japanese accent, it will be for CeVIO which has an English interface). Voiceroid and A.I.VOICE's interfaces are currently only in English (A.I.VOICE will be releasing an English voice in December, so they might make an English interface then). You can play back the audio by pressing the play button or export the audio as a wav file. I guess you could use it for real-time talking, but you would have to be okay with lags in conversation. Only some of the voice banks are available in 4 emotions (normal, happy, sad, angry).

Crypton has no speaking software, only singing. Synthesizer V is also singing only. They work through typing phonemes into MIDI notes and would not be useful for real-time conversation.