SynthV CeVIO AI / Synthesizer V Tsurumaki Maki

Rylitah

kiyoteru enthusiast
Staff member
Moderator
Apr 8, 2018
562
CeVIO's website has been updated with information about using English talk banks (i.e... just Maki). Specifically, it's about the English Talk editor.

The word "accent" is mentioned a bit. I'm assuming this is similar to Voiceroid and how when typing something like "こんにちはー" (konnichiwa), you will be able to edit each accent/syllable independently: "こ/ko" "ん/n" "に/ni" "ち/chi" "は/wa" and "ー/-". You can edit the pitch of each accent/syllable to put emphasis on any one of them. I've never used any form of CeVIO Talk in my life, but I'm certain this is what is being referred to when the word "accent" is used (also because Voiceroid also calls it accent).

So with that explanation out of the way, here's a brief rundown of what it says on the site (feel free to correct if anything is wrong):

- Half width alphanumeric characters are the only accepted input (cannot mix roman characters and Japanese kana/kanji)
- There won't be an "Accent Adjustment" screen when Maki releases. With Japanese banks you can edit the pitch of each individual accent (syllable) to be high or low like I previously mentioned; this won't be possible with English banks, but they plan on implementing an accent adjustment screen where you can adjust the "stress" of each syllable in three levels.
- Cannot split or combine accents (syllables) with English banks. This is a Japanese bank only feature.
- The English Talk editor has the "Word Registration," "Dictionary Management," and "Line Clip" functions. Can't use contents registered in English talk in Japanese talk and vice versa.
- You can register English words under "notation" (表記) or "reading" (読み), I'm guessing like registering abbreviations as saying the actual full word aloud. (I don't know what this part means: 「品詞」は汎用として登録され自動で適用されるため、「優先度」も設定不要です。; something about not needing to set the priority of generic part-of-speech terms because it's applied automatically? I don't know what's considered "part-of-speech" though)
- No "pause settings" for English banks. With Japanese banks, you can declare how long a pause is for each type of punctuation, but that's not possible with English banks. It mentions how in English, there isn't always a pause when a comma or period is used (such as in words like Mr. / Mrs. / etc.) so that's most likely the reason why.

At the end there's an image showing what editing the "stress" in three levels thing will be like, though there's a note saying it's under development and may differ from how it looks upon release. They plan on releasing it in August.

(Given that this news doesn't have an English translation though, I... am going to assume that there won't be an English version of the editor anytime soon...?)
 

peaches2217

The One With the Sunflower Hairclip
Sep 11, 2019
1,861
25
Arklahoma
I love everything I’ve heard from Maki’s English TTS so far, so I’ve got faith things will work out! This is the first time I’ve had faith in CeVIO AI, I’ll be honest 😅 but as Patuk said, she’s the first of her kind. She sounds perfectly natural in her demos, so I doubt any of the setbacks will hinder her too terribly much.
 

Rylitah

kiyoteru enthusiast
Staff member
Moderator
Apr 8, 2018
562
I only have the older CeVIO, I thought CeVIO AI was translated into English?
Is it? I've never used CeVIO in my life and always heard people complain that it never got an English UI even after IA released, but if it does have an English translation then disregard what I said about it ahah.

And yeah, it all sounded pretty bad on first read, but after thinking about it I think all of these things just make sense. The only thing that's really disappointing still is that there's no accent/syllable adjustment feature upon release and we'll have to wait at least two months after her release for it, but hopefully it'll be good when it does.
 

Leon

AKA missy20201 (Elliot)
Apr 8, 2018
808
Most of this is a bit of a bummer but ~ot too surprising for the first bank of its kind. But, does the accent thing affect homophones? I remember her one clip where she very clearly said the wrong "read". If you can't adjust that kind of thing on release, I think that's pretty yikes. But if it's only a 2 month wait, it's not such a big deal 🤔
 

Rylitah

kiyoteru enthusiast
Staff member
Moderator
Apr 8, 2018
562
Most of this is a bit of a bummer but ~ot too surprising for the first bank of its kind. But, does the accent thing affect homophones? I remember her one clip where she very clearly said the wrong "read". If you can't adjust that kind of thing on release, I think that's pretty yikes. But if it's only a 2 month wait, it's not such a big deal 🤔
The image showing how the accent adjustment feature will look like has each accent/syllable separated by phoneme, so it looks like you can change it...? I don't really know for sure, but I would hope it can be adjusted orz.

 

peaches2217

The One With the Sunflower Hairclip
Sep 11, 2019
1,861
25
Arklahoma
The image showing how the accent adjustment feature will look like has each accent/syllable separated by phoneme, so it looks like you can change it...? I don't really know for sure, but I would hope it can be adjusted orz.

I'm almost certain you'd be able to change it, seeing as how the phonemes are laid out as such. I think the only reason the "read/read" thing didn't get sorted out in her demonstration video is because, as with her previous CeVIO English demos, it was handled by someone/a team that speaks little English.

Realizing her team doesn't speak much English and she's still the most natural-sounding English TTS we've gotten from CeVIO... honestly, that makes her even more impressive to me. :maki_lili: :tsurumaki_lili::maki_lili: :tsurumaki_lili: :maki_lili: :tsurumaki_lili: :maki_lili: :tsurumaki_lili: :maki_lili: :tsurumaki_lili:
 

peaches2217

The One With the Sunflower Hairclip
Sep 11, 2019
1,861
25
Arklahoma
OK, I downloaded CeVIO CS7 & switched the interface to English so we can figure this out. (Note: Situation may be different for CeVIO AI, but I kinda doubt it lol)

The parts that aren't in Japanese are:
  • The Japanese text you write in for the character to say + the tuning of each phoneme written in katakana.
  • Presets for voices (vol, speed, pitch, mix of emotions, etc).
View attachment 4885

  • Talk > Japanese Word Registration & Dictionary Management popups aren't translated at all.
View attachment 4886
View attachment 4887


Techno-Speech has advertised the fact that they could make English, Chinese, and Japanese TTS for tons of years and they didn't even have the interface completely translated. I guess if you were to force me to defend them, maybe they expect non-Japanese speakers would need to register Japanese words.

I am highly curious to see if anyone who buys Maki English TTS would show us how easy editing her phonemes/intonation is and if it actually gives nice results. (I also can't wait to find out how A.I.VOICE compares with their interface for editing English phonemes in December.)
To tell you the truth, when I tried the CeVIO 6 trial a while back, I had no trouble finding my way around the interface and making Sasara say all manners of fun things, even as a filthy gaijin that doesn't speak much Japanese; I don't remember if it was translated or if it there were just multiple ways of doing it and I found the easier methods. Either way, I remember having a lot of fun with it, quality aside, and the thought of getting Maki's Talk banks is making me excited! If no one's published a brief tutorial by the time I get her I'd be happy to do so.
 

Leon

AKA missy20201 (Elliot)
Apr 8, 2018
808
The image showing how the accent adjustment feature will look like has each accent/syllable separated by phoneme, so it looks like you can change it...? I don't really know for sure, but I would hope it can be adjusted orz.

OK, then that sucks. Because what I was getting at with my original question was, if this feature won't be available for her English when it first launches, you can't change it if she says the wrong word? Oof. But I imagine not that many things will be released in the first two months, so it would be fine for people who buy/use her after.

And yeah! @peaches2217 I figured the read/read was because it wasn't native speakers making demos, but I wanted to check to make sure that could be changed at all. Seems like yes, but not right away. Ah well.
 

peaches2217

The One With the Sunflower Hairclip
Sep 11, 2019
1,861
25
Arklahoma
What sets Voiceroid, AI voice, Cevio talk/etc. apart from all these regular online tts's is the ability to fine tune what's being said and I don't see how that being more complex and time consuming in english language, which is phonologically very complex, is exactly unreasonable.
I think that's what a lot of people are discrediting: this is AHS' first venture into English TTS, and for a maiden voyage, it's going really well. IA's the only English TTS we've had on CeVIO thus far, and she didn't receive nearly as much attention, publicity, or marketing as Maki's getting, so it stands to reason things would start off a little shaky. Were this coming from an engine like Vocaloid or SynthV, it'd be another story, but for an engine that's been pretty Japanese-exclusive? I think it's impressive. And I'm not one to speak in defense of CeVIO AI. :clara_ani_lili:
 

Rylitah

kiyoteru enthusiast
Staff member
Moderator
Apr 8, 2018
562
IA's the only English TTS we've had on CeVIO thus far, and she didn't receive nearly as much attention, publicity, or marketing as Maki's getting
IA doesn't actually have an English TTS - her talk is only for Japanese, and that includes her recent update. Maki's the first English TTS we're getting from any of these companies ever, so it's even more understandable that there'll be some things to iron out (it is just a bit odd that such an important feature won't be there at launch, but that really has nothing to do with Maki herself).

I think it all boils down to what you really want to use these synths for. In the end, Maki English, and literally every other Japanese company-made English synth (talk or otherwise), is and always was going to be a niche product. A lot of us hobbyists like fine-tuning things and getting that most perfect result (see: tuning singing synths), so having to do that here won't be a problem. Alternatively, it's okay if you don't want to do all that too; if you don't like Maki's accent and don't want to fine tune everything, then you don't have to buy her. I think there's reason to be critical of the editor too, of course - but if Maki does well and more companies see a market for English TTS, then it can only be a good thing.

I think it's cool that people are genuinely getting interested in talking synths now; as someone who's used VOICEROID since 2016 and always wanted to see the market expand, this has been my dream, haha.
 

peaches2217

The One With the Sunflower Hairclip
Sep 11, 2019
1,861
25
Arklahoma
IA doesn't actually have an English TTS - her talk is only for Japanese, and that includes her recent update.
Motherfucker I'm an idiot. I coulda sworn she had an English talk bank. Shows you how much I know. 😅 Forgive me!

But I'm with you here. She's a niche product, and while I understand many criticisms about her, criticisms like "She doesn't sound like a native speaker!" just make zero sense to me whatsoever. Like... no, she doesn't. That's the point. If they wanted her to sound like a native speaker, they'd hire a native speaker. If you don't like the fact that she sounds non-native, rather than moan about how tragic it is that her VP can't perfectly imitate an American accent, just... don't buy her. That's where all of my frustration comes from; all other criticisms I've seen I either agree with or can understand completely.

I love Maki, so I wanna remain optimistic for her and CeVIO AI's future! And I want more products with her quality on CeVIO AI. After the disappointments of Yukari Rei and Rikka's talk bank, this is a breath of fresh air, even in its unfinished state. And as you said, people are finally getting on board with talking synths, and that'll just usher in a new era of varied, quality, well-funded TTS banks, as interest in singing synths did to Vocaloid in the late 2000s/early 2010s.

Ah, I still remember hearing Heartbreak Headlines and being like "Holy shit, that's a talking Vocaloid! That's amazing!" Seeing how far it's all come nearly ten years later is making me nostalgic :avanna_lili:
 

Users Who Are Viewing This Thread (Users: 0, Guests: 0)