I’m genuinely unsure if there was ever a dedicated thread for these vocal updates in general. I’ve seen them here and there within the Cross-lingual Singing Synthesis announcement thread, but that’s about it. I figured starting a thread all in one place just for the latest development and demos of these versions would be good to have.
This is about the AI voice databases for the other 6 characters from Quadimension (Beijing Photek) that are currently available as Standard only. For a bit of background, these were announced since December 23, 2021. Details included that the 6 would be updated especially to support Japanese and English due to Cross-lingual Singing Synthesis being implemented in the 1.5.0 update and the only way the 6 could use that was if they had AI type databases since Standards couldn’t use it. These AI databases were noted to be trained by the Standard versions. The vocals are currently going through beta testing to ensure satisfactory results.
As of now, only Cangqiong and Muxin have examples shown in public, with both being labelled as “Synthesizer V Plus”, which is referring to these AI versions. Here’s a list of the demos and samples for both of them, both directly hosted on the official MEDIUM⁵/Xingchen/Haiyi accounts and off (only posted to the beta testers’ accounts). I’ll skip the short versions if they ended up getting full ones and post them in release order.
“Royals” short cover by Creuzer ft. Cangqiong Plus (English Crosslingual)
Original by Lorde
" 孤勇者" (The Lonely Warrior) cover by 瑞安Ryan ft. Cangqiong Plus & Muxin Plus (English Crosslingual & Chinese)
Original vocalist: 陈奕迅
Original music: 钱雷
Original lyrics: 唐恬
English translyrics: 肖恩Shaun-Gibson
“Into the Unknown” cover by 瑞安Ryan ft. Cangqiong Plus & Minus Standard (English Crosslingual)
Original vocals: Idina Menzel
“One last kiss” short cover by 佑可猫 ft. Muxin Plus (Japanese & English Crosslingual)
Project file: Selio
Original vocals: 宇多田光
“Someone Like You” cover by 瑞安Ryan ft. Cangqiong Plus (English Crosslingual)
Original vocals, music, & lyrics: Adele
Original music & lyrics: Dan Wilson
“Solar Storm” cover by 佑可猫 ft. Muxin Plus
SVP file: 尤乌鱼
Original music: Li Zong
Original lyrics: 大古
Original vocals: Xingchen
“Let It Go” (25 language version) cover by 瑞安Ryan ft. Cangqiong Plus
Original vocals: Idina Menzel
“Lovely” cover by 佑可猫 ft. Muxin Plus (English Crosslingual)
Original by Billie Eilish & Khalid
Things don’t seem to be looking too good for these Plus versions. A fan asked if these would be releasing soon and Quadimension responded that it’s probably not likely. The response was given on September 29 this year.
Hi everyone, I’m currently doing massive edits on SynthV wiki, so I’m just popping in real fast about what was recently revealed.
On an article released by Ddickky, one of the fans asked in the comments about the status of the Plus voices and Ddickky basically responded that these databases were created to demonstrate the effects of training AI with Standard databases. These were not meant to be released to the public and were only sent to shareholders to demonstrate the technology, not as a product update. Because of the way they were produced, they couldn’t be called Synthesizer V AI voice databases as this would affect the brand quality of AI voice databases, hence why they’re known as Synthesizer V Plus.
Direct comment here:
Unfortunately, we don’t know if they’ll pursue full AI voice databases in the future or not, or about their next projects for Synthesizer V in general if any (not that was pointed out to me nor what I can read).
The standard-trained AI voice databases were promoted with Xingchen Infinity. The VAs were invited at the same time in the prior year.
There were some concerns about the AI voice databases sounding too close to the VAs. The VAs were concerned that their jobs would be “stolen”. There was also mention of procrastination on their end as well until more recently. Basically, there’s been some reluctance to cooperate in producing AI voice databases.
There’s concern that the AI versions would not sound like the R1 and R2 Standard versions. For additional context, there was some resistance to Xingchen Infinity because she didn’t sound like her original VOCALOID4 version. Xingchen needed to be updated because the V4 version performed poorly, so it needed to be launched even under the pressure of changing the singing voice style. For the other cast members, changing the voice style was always uncertain. There may be reactions of resistance similar to what Xingchen Infinity had gone through.
There was a lack of motivation due to commercial sales of voice databases never being so profitable.
Regarding commercialization, there’s no solid plan right now, but it might be similar to Xingchen Infinity. Because the tech workload and input costs were different, they may be sold for hundreds of yuan, which was cited to be another reason that he thinks that they might not be suitable commercially.
The good news? We are probably seeing the production of Chiyu AI and Minus AI, both expected for this year if there are no delays and/or issues.
Reminder that the Plus versions should not be expected. These are for private use.
Currently, no mention of Shian, Cangqiong, Haiyi, and Muxin.
Let me post another article today. It’s better for everyone to be clear that: according to the information I got until now, Chiyu AI is definitely going to be produced, my family, friends and me myself are always ready to work with it. The progress is slow because we’re very busy and because there are some other considerations in the voicebank. We need to make a demo version of voicebank, and adjust the feature and quality accordingly before publish the final version.
In my point of view, the AI version may not sounds like Youmao (UNEKO) that much as former versions did. There exists lots of praise about the AI voice databases sounding so close to the VAs, but nobody can afford the risk to publish another version so similar to VA. BUT, the success of Mai may show another road, that is to clone VA’s as-is. As for me, if(and only if) the high part of Chiyu performs well, both way is acceptable.