• We're currently having issues with our e-mail system. Anything requiring e-mail validation (2FA, forgotten passwords, etc.) requires to be changed manually at the moment. Please reach out via the Contact Us form if you require any assistance.

Other Unreasonably long post about Medium⁵ Haiyi and Cangqiong's statues as virtual singers

xuu

long suffering synth fan
Apr 8, 2018
671
23
UK
A few people were wondering about how I'm so sure that Medium⁵ Haiyi and Cangqiong are UTAU, so I decided to write up a post about it. :haiyi_lili:


As a foreword, about how they are officially described. What is public knowledge is that they are Chinese virtual singers, developed by Beijing Photek S&T Development Co., Ltd., the business arm of Chinese music circle Quadimension for an undisclosed engine, which is a fancy way of saying they're not VOCALOIDs. They are part of a group called Medium⁵, of which the members are Xingchen, Haiyi, Cangqiong, Shian and Chiyu, in order of reveal. Quadimension uses these voices for their original music and release albums and merchandise with them.


“Why did Quadimension keep Haiyi private?”, you may ask. Ddickky, the head of the group, said it was due to costs of “the fee for distribution”, and also because of the fear of cracking and illegal distribution.


Personally, I assume that he is referring to the fee Yamaha asks for when a company wishes to release a VOCALOID. While it costs nothing to develop a VOCALOID if you have the development kit, you need to pay the steep fee to have Yamaha approve the character for release. This rules out VOCALOID as the engine for either Haiyi or Cangqiong from the get-go, as Quadimension are basically saying they don't make enough money from sales to justify it, especially considering they probably paid a decently large sum to Shanghai HENIAN for them to develop Xingchen for VOCALOID4.


To add to the cost argument, Haiyi has four DBs that have been revealed: Normal, Dark, Power, and Aroma. Given that Yamaha charges per DB, it would have cost Quadimension at least four times what it would have cost them to release all four on VOCALOID. They most likely decided to go for another synthesizer so they could have more expressions to use in their music.


So, what about other synthesizers?


The synthesizers I see guessed the most other than VOCALOID and UTAU are Sharpkey, CeVIO, and Synthesizer V. I was personally in the Sharpkey camp for Haiyi and was expecting Cangqiong to be a Synthesizer V vocal, but I'll get into that later.


Starting with CeVIO since it's easy to get out of the way, CeVIO is programmed through machine learning in association with the Nagoya Institute of Technology. When I say machine learning, I mean on supercomputers. I learnt this while being peer to the development of another synth vocal who potentially could have been on CeVIO. It takes on average 4 months or so for the process to be completed, which is why it takes so long between vocals. Midorizaki Kasumi was used in 2013/14 for example, but they had to wait for the other vocals (I can't find a link to the post any more, so you'll have to trust me on it). Due to this, development of 4 Chinese DBs when there had been none before would have probably taken well over a year, which doesn't fit into Haiyi's timeframe, assuming she started development around mid-late 2017.


Next, to Sharpkey. I know a few here are users of Sharpkey so they will understand when I say Boxstar is a very busy man. He was until recently the only developer working on Sharpkey, and certainly has been the only one programming DBs, as well as working a day job. As it was noted that Haiyi's DBs were completed around Janurary 2018, we'll be using that timeframe. Around this time, Lian Hua and Yong Qi were released. I'm not sure about Feng Qing Wakasame since I don't really know anything about him, but he very well could have been worked on too. It's highly doubtful Boxstar would have had the time to work on four DBs for one character at that time, ruling it out. I mean, it's possible that there might be a Medium⁵ vocal on the software in the future, but that isn't Haiyi.


Finally, Synthesizer V. The software itself was revealed in October 2017, though no names were actually revealed until August 2018 when Eleanor Forte was revealed. Kanru Hua has stated that even Eleanor was only intended to be a test vocal for the engine, and thus they signed contracts saying they could not disclose her voice provider's name, or any personal details. He worked on this vocal alongside JP-F1, MAN-M1, and MAN-F1. This makes it very unlikely Haiyi would be a Synthesizer V vocal, though she generally does not sound like one anyway.


This leaves either UTAU or some other synthesizer I have neglected to mention, and leads into why I'm sure she, and by association Cangqiong, are UTAUs.


As you can see in their demos and highlight reels for albums, all of the songs featuring Haiyi and Cangqiong are tuned by NAPSTINA, a well-known UTAU tuner in China. I haven't done much research into them but as far as I am aware they do not use other synthesizers often. I'm also vaguely aware of Beijing Photek hiring Haru.jpg, arguably the best person in China for oto.ini configuration, around the time of Haiyi's announcement. Those two are the only concrete things that I have that scream “IT'S UTAU” to me, but in terms of intangible things...


They just, sound like UTAUs? You know, when you've used something a bunch and you end up developing a knack for telling what software/resampler a vocal is using, unless it's very well masked. Haiyi and Cangqiong are without a doubt extremely high quality vocals, who were near certainly recorded in the same conditions as VSINGER vocals are. This means they don't exactly sound like the average UTAU, but transitions can give it away. I must give it to them, they're configured extremely well, but we all know a few UTAU that are done so well they almost sound human. They also lack the qualities I personally associate with VOCALOID/CeVIO/Sharpkey/Synthesizer V, which I know people love to refer to as “engine noise”.


I know I basically only spoke about Haiyi, but everything I said applies to Cangqiong, though it seems she only has one DB at the moment. In UTAU, Beijing Photek have the freedom to record as much or as little as they want, all while keeping them as “private vocals”. They would probably sell them if they could, but UTAU vocals are super easy to redistribute, as everyone and their mother can tell based on how many people conveniently have Matsudappoiyo and Matsudappoine years after they were retired and taken down.


If you have any questions or want me to clear something up, feel free to ask! Sorry it's such a long post.
 

uncreepy

👵Escaped from the retirement home
Apr 9, 2018
1,618
Thank you for taking the time to explain your theory on why you believe they are using UTAU.

A thing I noticed is when you click the Cangqiong demo , it is labeled as "VOCALOID·UTAU". There are other labels like MUTA and even Synth V, so if it was Synth V, wouldn't it be tagged properly? Also, Kanru Hua has been working on Synth V and giving us teasers all along the way, and none of them sound like Cangqiong. I don't think he would secretly develop her voice, because he's been quite transparent about the process of making Synth V's current voices.
 
  • Like
Reactions: ngoomie and xuu

xuu

long suffering synth fan
Apr 8, 2018
671
23
UK
Thank you for taking the time to explain your theory on why you believe they are using UTAU.

A thing I noticed is when you click the Cangqiong demo , it is labeled as "VOCALOID·UTAU". There are other labels like MUTA and even Synth V, so if it was Synth V, wouldn't it be tagged properly? Also, Kanru Hua has been working on Synth V and giving us teasers all along the way, and none of them sound like Cangqiong. I don't think he would secretly develop her voice, because he's been quite transparent about the process of making Synth V's current voices.
Yes, it would be tagged properly. On Bilibili, if you want VOCALOID fans to see something you'd tag it with "VOCALOID·UTAU" anyway, so people weren't too sure if it was or not. It's the best default tag if you don't want to disclose an engine.
Regarding development, it would probably be Animen or Beijing Photek working on it themselves, for example other a demo each we had nothing really for GENBU and AiKO until their release. The voice is pretty clearly not SynthV anyway. In regards to ones Kanru is still working on, we can expect JP-F1 and MAN-F1. MAN-M1 was cancelled.
 
  • Like
Reactions: ngoomie

ngoomie

New Fan
Jan 22, 2019
19
If they are UTAU voicebanks, what resampler do you think they might be using in the demos? I can't really hear anything in the Cangqiong demo demo that points to her being an UTAU, what I sort of heard from that demo was maybe Sharpkey, but I also haven't used Sharpkey in a long while, so my memory of what it sounds like is probably fairly off. The only resamplers I think I've really heard super recently have been the default resampler, moresampler and tn_fnds, so that also might be why I don't really hear UTAU when I hear that demo.

EDIT: the more I listen to that demo, the more I can hear UTAU, so I'm definitely leaning more towards that now. I'm still curious about which resampler you think is being used though so I can compare with other examples of it.
 
Last edited:

xuu

long suffering synth fan
Apr 8, 2018
671
23
UK
If they are UTAU voicebanks, what resampler do you think they might be using in the demos? I can't really hear anything in the Cangqiong demo demo that points to her being an UTAU, what I sort of heard from that demo was maybe Sharpkey, but I also haven't used Sharpkey in a long while, so my memory of what it sounds like is probably fairly off. The only resamplers I think I've really heard super recently have been the default resampler, moresampler and tn_fnds, so that also might be why I don't really hear UTAU when I hear that demo.

EDIT: the more I listen to that demo, the more I can hear UTAU, so I'm definitely leaning more towards that now. I'm still curious about which resampler you think is being used though so I can compare with other examples of it.
I think it's probably some iteration of fresamp. With the right configurations it can sound really good. It could also be something like model4, for example, or maybe moresampler with some configuration changes? My best bet would be on fresamp though because it's always what people are using in really good samples. I definitely hear no Sharpkey and I've beta tested a couple of them.
 
  • Like
Reactions: ngoomie

Ulysses

from VOICeVIO
May 4, 2018
40
(We can skip the possibility of CeVIO as long as I'm watching on C family. There is only one Chinese company collaborating with CeVIO.)
I didn't focus much on Chinese virtual singers but at least I read Chinese news.
CangQiong might be UTAU since the tuner of a demo song seems to be a pro UTAU user. However, since Beijing Photek is now the official sale agent for SynthV in China (mainland), the following voices (after HaiYi and CangQiong) by Quadimension are likely to be SynthV voices.
 
  • Like
Reactions: xuu and ngoomie

uncreepy

👵Escaped from the retirement home
Apr 9, 2018
1,618
What if they start selling Haiyi and CangQiong on Synth V, just like Renri used to be an UTAU but is now on Synth V? It would be nice if Haiyi fans could finally use her voice for their own songs.
 

xuu

long suffering synth fan
Apr 8, 2018
671
23
UK
What if they start selling Haiyi and CangQiong on Synth V, just like Renri used to be an UTAU but is now on Synth V? It would be nice if Haiyi fans could finally use her voice for their own songs.
It would be cool to see... and bad for my wallet. I think it's possible? I'd definitely expect for Shian and Chiyu to be SynthV now. Glad we can always rely on @Ulysses~~
 

DefiantKitsune

Lonely kanon fan
Apr 11, 2018
621
A quick check on Vocadb shows Napstina is not credited for anything but songs involving Haiyi, Cang Qiong, and Stardust.

EDIT: Now that i've posted this, I'm not sure how it's relevant
 
  • Wow
Reactions: xuu

YunīC

Stan Yuezheng Longya
Apr 8, 2018
211
hmmm.... for Haiyi Utau defenitely seems likely.... for Cang Qiong I kinda feel like it could be SynthV as well, since Utau & SynthV sometimes sound a bit similar.... (though Cang Qiong does sound a bit more HQ than SynthV but then again I she's the first with such a voicetype so maybe it's just not so noticeable with that...)

But actually it doesn't matter because some companies have done it before to release demos with utau and release the vocal on another engine coughAlyscough
(In truth I just don't want Cang Qiong to be utau because I really do not want her to stay private since I love her and would like to get her one day)
 

xuu

long suffering synth fan
Apr 8, 2018
671
23
UK
hmmm.... for Haiyi Utau defenitely seems likely.... for Cang Qiong I kinda feel like it could be SynthV as well, since Utau & SynthV sometimes sound a bit similar.... (though Cang Qiong does sound a bit more HQ than SynthV but then again I she's the first with such a voicetype so maybe it's just not so noticeable with that...)
It's definitely not SynthV. It wouldn't sound so... strong? And her transitions don't sound like it at all. I hope that they eventually port or rerecord them for SynthV though because I also really want them!
 
  • Like
Reactions: YunīC

YunīC

Stan Yuezheng Longya
Apr 8, 2018
211
Oh yeah I forgto to say so but on the 2nd Song she really sounds way more like utau than on the first which makes me so sad because I do not want her to be private :(
 
  • Sad
Reactions: xuu

Users Who Are Viewing This Thread (Users: 0, Guests: 1)