A.I.VOICE (AITalk5) General News

[uncreepy]

Link to the PDF: https://ssl4.eir-parts.net/doc/4388/tdnet/1751595/00.pdf
From AI Talk's news section (September 13th): 株式会社AI（エーアイ）

(Thanks to Fumito Fumizuki https://twitter.com/fumito_fumizuki for pointing this news out in the Discord.)

I translated the important parts:

AITalk is currently at version 4 (AITalk®4). The temporary name for the new engine is AITalk®5.
Because the vocal synth scene changed from mainly being used for robo calls and learning videos to now being used for smart products with interactive capabilities, they got a grant during July 2017 to December 2018 to develop a deep learning synthesis engine to solve issues with the old engine.

AITalk 4 uses "corpus-based speech synthesis": To meet the needs for interactivity, the voices require emotions like happiness, sadness, anger. Corpus-based speech synthesis needs both a phoneme dictionary and rhythm dictionary so it can figure out the accent. But it cost a lot and transitioning between emotions is not smooth.

The next gen AITalk 5 uses deep neural network speech synthesis. The sound quality will go up, the pronunciation will sound more natural/human, and it will naturally switch between emotions such as joy, sadness, anger. Instead of having an emotion dictionary like corpus-based speech synthesis had, the deep-learning one will have less files to record and therefore be less expensive.

The new engine will be available April 2020.

For those who don't know, products such as GynoidTalk and VOICEROID use AITalk.

 

[Prism]

I hope it sounds like the new cevio demo. Google's wavenet (ai tts) has some engine noise and banks are inconsistent. With Amazon's polly (has both ai and non ai tts) I actually prefer the non ai most of the time it just sounds clearer. Excited to hear about a demo

 

[uncreepy]

Note:
This page talks about SDK stuff, probably don't need to read it, but I'm leaving the link here anyway.

AiTalk5.0 has a page where you can hear samples!

音声合成ソフト、読み上げ、人工・電子音声なら【株式会社 エーアイ(AI)】

次世代型音声合成エンジンAITalk5は、従来の 「コーパスベース音声合成方式」と、最新の深層学習技術を活用した「DNN音声合成方式」を、利用シーンに合わせて選択することで、さらなる人間らしさ・豊かな音声を追求した高品質音声合成エンジンです。

www.ai-j.jp

Scroll down and there are 12 sound clips. All 12 clips are using their character named Nozomi, who has always had 4 emotions (normal, happy, angry, sad) (she's the 1st girl on this list). The 12 clips are just various stuff like phone call examples, train announcements, automated payment messages, etc (the script is to the right of the sound button).

As a side note, I am not sure if Nozomi is actually a currently existing Voiceroid (still not that good at recognizing the voices of the ones I don't care too much about). For example, Sumire is confirmed to be rebranded Yuzuki Yukari.

 

[Rylitah]

I can't tell much of a difference, ahah ;; Oh well. Future Voiceroids will probably be on this engine, so that'll be something fun, even if I personally don't hear much of a change.

As a side note, I am not sure if Nozomi is actually a currently existing Voiceroid (still not that good at recognizing the voices of the ones I don't care too much about). For example, Sumire is confirmed to be rebranded Yuzuki Yukari.

She's not.

On the demonstrations page, "Maki" is Tsurumaki Maki, "Reina" is Kotonoha Aoi, "Taichi" is Minase Kou, "Sumire" is Yukari like you mentioned, "Anzu" is Tsukuyomi Ai, and "Koutaro" is Tsukuyomi Shouta. The rest aren't Voiceroids.

 

[uncreepy]

I am wondering if the not very noticeable difference is just because Nozomi was high quality already? Maybe you can better judge the naturalness for lines that are more... real human dialogue-y (as in not just a calm train announcement like one of the clips and more like Let's Play dialogue-y). I wonder if the interface could be easier to use or tuning wouldn't take as long to tune with the deep learning feature? Saving time would be a nice perk compared to the current version.

Thanks for telling me which character is a Voiceroid, I was dying to know!

 

[Prism]

Am I the only one that is a little bit disappointed. There's so many new more natural ai tts that are on the market and this is a little disappointing in comparison.

 

[uncreepy]

Yeah... I guess it's hard to know for sure when we only got to hear 1 voice, it didn't show off the range of emotions, and we didn't get to see the interface. Hopefully if a vocal synth company like Voiceroid upgrades their version from 2 to 3 (or something) then we will know for sure what it can do. I don't know if other TTS software has such detailed sliders for speed, intonation, pitch, etc, so if new AITalk has that in combination with deep learning, I think it would be quite good.

 

[uncreepy]

kadotanimitsuru tweeted about AITalk5 plans being released today. They are going to sell an original brand of products aimed at individuals. The planned release dates are as follows:
1) Feb 2021
2) July 2021
3) December 2021
He thinks it's probably a rehauled Kantan! AITalk, but the press release actually says the brand name is undecided.

Reading over the notice, it says that they want to create convenience and fun with speech synthesis. They want to do this by selling under their own original brand using the AITalk5 technology and are going to expand to from BtoC (business to consumer) and not just BtoB (business to business).
They haven't decided the product name yet. They want to expand globally and want to recruit other companies which own characters so they can increase their character lineup.
They expect the impact on sales to be small at first, but gradually increase.

 

[uncreepy]

I didn't notice it, but a notice went up on the homepage on Aug 27th saying the same thing but aimed at customers: エーアイ、オリジナルブランドによるAITalk®5 個人向けキャラクター音声読み上げソフトを2021年2月リリース予定。 株式会社AI（エーアイ）

It says AITalk5 will be released under an original brand Feb 2021 aimed at individuals. They feel it is necessary to expand their business to include planning, development, and sales for this new series.
They want to increase the current character lineup, work to maintain the characters, expand globally, and are aiming for the entertainment sphere. They will announce details on a website "soon".

I noticed that they have a section to email them relating to PR. I wonder if it would be OK to send them a message asking questions about it? Like if it would be something similar to what VOICEROID is with cool characters or if they want to make an English voice for us. What do you think? Any questions you'd want me to ask if I did write an email?

 

[uncreepy]

I forgot to mention that any videos from 5 months ago (as in posted in April) are using the new engine (so the narration in the following clips are all AITalk5):



In the last clip, it looks like you can still change things like speed/intonation/etc.

 

[uncreepy]

It seems that a website for the new brand for AITalk5 has been made public on September 30th. It will be called "A.I.VOICE".

Announcement: エーアイ、2021年発売のオリジナルブランドによる個人利用向け音声読み上げソフトの名称を「A.I.VOICE™」に決定 株式会社AI（エーアイ）
Website for A.I.VOICE: A.I.VOICE™ - ティザーサイト

It is set to debut in February 2021. They say to please follow the teaser website for future announcements. The website touts the naturalness and power of expression that AITalk5 is capable of. A list of detailed functions for the software and characters is listed as "coming soon".

 

[uncreepy]

Didn't have time to write about this before, but on the 29th, it was announced they are recruiting companies who want to have consultations to find out how to make their character more famous by turning them into a synthesized narration product, how much it will cost to make an AI VOICE bank with them, find out the amount of profit they can make from that, what they will need to provide in order to achieve this, etc.

 

[uncreepy]

A.I.VOICE is going to get Chinese and English voices in December 2021. (What a long wait!) The weird part of this article is that, in addition to an A.I.VOICE, it also specifically mentions a 歌声 (singing voice) as well. So they must be planning to market characters that can both speak and sing.

Source: エーアイ、個人向けオリジナルブランド製品「A.I.VOICE™」による歌声合成ソフト・外国語音声合成ソフトを2021年12月リリース予定 株式会社AI（エーアイ）

 

[xuu]

It's making quite a bit of noise on my Japanese TL from what I can see, with a lot of reference to this article. Very spicy, and I'm looking forward to hearing it. Seems like the synthesis battlefield will widen even more and it will become a competition between CeVIO, Synthesizer V and A.I.VOICE...

 

[lIlI]

Voiceroid is getting a singing counterpart, huh?

 

[uncreepy]

Kotonoha Akane and Aoi and Iori Yuzuru have been announced to be getting voices for A.I.VOICE February 2021. They will gradually be announcing which characters are next (in the past, they said the release dates would be February, July, and December 2021).

Source: エーアイ、個人向けオリジナルブランド「A.I.VOICE(TM)」より 「琴葉茜・葵」「伊織弓鶴」を製品化 [ まるごと広報代行サービス PRナビ ]

I assume that these characters will get both a song and talk voice. I already own Yuzuru, but I wouldn't be opposed to buying him again or getting his song voice if sold separate.

 

[Blue Of Mind]

Kotonoha Akane and Aoi and Iori Yuzuru have been announced to be getting voices for A.I.VOICE February 2021. They will gradually be announcing which characters are next (in the past, they said the release dates would be February, July, and December 2021).

Source: エーアイ、個人向けオリジナルブランド「A.I.VOICE(TM)」より 「琴葉茜・葵」「伊織弓鶴」を製品化 [ まるごと広報代行サービス PRナビ ]

I assume that these characters will get both a song and talk voice. I already own Yuzuru, but I wouldn't be opposed to buying him again or getting his song voice if sold separate.

If Yuzuru gets a song VB, I will freak the hell out because I loved his voice type as a talk bank, and I was really sad he wasn't a Vocaloid or anything.

 

[Rylitah]

February 21 release date, pricing details (including early preorder prices starting in January), boxart illustrations for the Kotonohas and Yuzuru revealed on this page, among other stuff

(Would add more but I'm about to sleep and the picture sizes rn are for ants; cute from what I can see though)

edit:

 

[denev]

A demo of the two was shown on the birthday stream. I recorded it and uploaded it here!
AI Talk 5 Demo

 

[Rylitah]

HOLY shit Yuzuru caught me so off guard. I'm a bit iffy on the Kotonohas; not sure if I like that tone, but they definitely sound less "text to speech" than their Voiceroid -- but I like their Voiceroid tone more orz. Was the new bank (蕾) used? Maybe the old tone is still there if so.

Yuzuru though....... wow. Sounds exactly the same, but lost a lot of that AI Talk noise and sounded almost human (there was still a little bit of it there, but only in one small part) - I wish they could've played a bit more of him!