According to official sources, this project was announced to currently be in production or is in its conceptual stages.
Please read the page for more information.
Applies to: AI voice database
This article is about the Synthesizer V Studio software known as a voice database. If you are looking for the Synthesizer V character then click here.
History[]
2020[]
In January 2020, Ddickky confirmed that he learned about the second generation of the Synthesizer V engine in mid-2019. He wrote that it was in the works and was expected to be ready by the end of 2019, but it was pushed to around February or March 2020. He confirmed that Haiyi was the last to release for the first engine and that the next voice database would release for the new one. In addition, he had plans to gradually update each of Beijing Photek's first engine voice databases, and depending on the development costs, users who owned any of the four vocals would be able to receive the update for free or for a low price.[1]
On April 9, Chiyu's "SVR2" version was announced to be revealed in addition to information about the updated engine itself.[2] On June 26, it was announced that Chiyu and the other Beijing Photek Synthesizer V vocals would receive a ported update to Synthesizer V Studio. These would be released on July 10. Users who already owned Synthesizer V and/or the original MEDIUM⁵ vocals received a special discount.[3]
On July 30, Chiyu was made available as a Lite voice database, which was a free version of her voice library that was ideal for trying her out.
2021[]
On December 23, 2021, it was announced that Chiyu, Shian, Cangqiong, Haiyi, Muxin, and Minus were expected to be upgraded with AI voice databases and would be able to support Mandarin Chinese, Japanese, and English (in correspondence to the Synthesizer V Studio 1.5.0's Cross-lingual Synthesis capabilities). It was clarified that the AI versions were trained by the Standard voice databases, and were not meant to be replacements for Standard nor would the update be available directly via the editor. As a separate type of voice database, the new libraries would need to be repacked in new installation packages, which required separate activation codes.
Ddickky commented that because Christmas was approaching and that Dreamtonics Co., Ltd. would be on holiday, it was thought that the voice databases could all be shipped on December 23 at 17:00 to directly align with the 1.5.0 engine update; however, he found this to be wrong. After Dreamtonics finished packaging the AI libraries, the activation codes were planned to be distributed by the Taobao agent, where users were to approach them, have their proof of purchase verified, and a new AI code would be sent. This was an entirely manual process and would have been done one by one. In addition, considerable time would have been needed to complete all six packages, so only one of them would have been finished on the 23rd, while the others would be finished later on.[4][5][6][7] The upgrades would be distributed to existing users for free and prices would be raised for new buyers soon due to the new inclusions.[8][9]
Later on, Quadimension commented that the release of the six AI voice databases was canceled for December 23 due to a lack of confidence about their performance as rushed products. A limited beta-testing would run first. The voice databases would be released after their performances were proven satisfactory. By then, Quadimension hoped to find an efficient way to distribute the AI activation codes.[10] On December 30, the voice databases were officially called "Synthesizer V Plus".[11]
2022[]
Throughout 2022, Cangqiong Synthesizer V Plus and Muxin Synthesizer V Plus received multiple examples demonstrating their capabilities as Standard-trained AI voice databases, including the Cross-lingual Singing Synthesis function. However, news about the development and future releases in general for all six characters were not available. On September 29, in response to a fan's question about whether the Plus versions of the six voice databases would be released soon, Quadimension wrote that it was probably not likely due to the qualities being unfit for commercial use.[12] On November 22, in response to a fan, Ddickky confirmed that the Plus series of voice databases were created with the intention of demonstrating the effects of training AI with Standard voice databases and were not meant to be released. The voice databases were sent to shareholders as a technology demonstration and not as a full product update. As these voice databases were to show the effects and technology of splicing AI training, they could not be noted as "Synthesizer V AI" as they would affect the brand quality of AI voice databases, thus the need of calling these "Synthesizer V Plus".[13][14]
2023[]
On March 11, 2023, in response to some fans' questions regarding the production of AI voice databases, Ddickky said that the other "AI" voice databases were promoted with Xingchen Infinity and that the voice providers were invited at the same time in the prior year. He said that he had no issue with the funding, but that some of the voice providers were concerned that the AI would be "stealing their jobs" while some were procrastinating with the work until recently. There was also a concern that the databases would not sound like the previous R1 and R2 "Standard" versions. He explained that an update for Xingchen V4 was needed as it performed poorly, so it must be launched even under the pressure of changing the singing voice style. For the other characters, however, the voice style change was always uncertain and it may create a reaction of resistance similar to Xingchen Infinity. Regarding the commercialization of the AI voice databases, Ddickky noted that the commercial sales of the voice databases have never been so profitable, so there wasn't enough motivation before. The commercialization would be determined after the production was completed, but it was expected to be handled similarly to Xingchen Infinity and would be sold directly since the technical workload and input cost were completely different. The AI voice databases may be sold for hundreds of yuan, which was one of the reasons why he thought they may not be suitable commercially. He also said that it was not a good time to speak from the perspective of business strategy and can only make a reminder in two months rather than rushing to ask every day. If there were no issues or delays, Chiyu AI and Minus AI were expected for 2023. Ddickky reminded fans that because there was no product line for the Standard-trained AI voice databases, they should not be expected for release.[15][16][17] On March 31, YouMao, Chiyu's voice provider, noted that as far as she knew, the AI voice database would definitely be completed.[18]
2024[]
On December 8, 2024, when asked by a fan in response to the announcement of Haiyi receiving a full AI voice database sometime in 2025, unofficial news outlet SynthV News responded that Chiyu AI had sounded too much like YouMao, leading Quadimension to respect the voice provider's wishes and postpone the voice database indefinitely, with a note that she "most likely" would be "re-recorded at [a] later date".[19]
Voice Database Information[]
Demonstrations[]
Synthesizer V Studio Standard[]
| Demonstrations | |
|---|---|
|
| |
| 404 Not Found |  YouTube  bilibili
|
| New Era of Virtual Singers Music Library[20] | |
|---|---|
|
| |
| 404 Not Found (2022 Version) | bilibili
|
Synthesizer V Plus[]
There were no demonstrations for Chiyu Synthesizer V Plus.
Synthesizer V AI[]
There were no public demonstrations for Chiyu AI, in light of her sounding too much like her voice provider YouMao and being postponed indefinitely.[19]
Voice Databases[]
- Regarding the Lite version:
- Chiyu Lite may not be used commercially in any way
- When using the Lite version of this voice database, "This work uses the Lite version" must be mentioned on any uploads.
- Chiyu Standard Solfège:
- Chiyu Standard Lite Solfège:
- A preview of Chiyu's Synthesizer V Studio Standard voice database (provided by the Dreamtonics website):
- This would have been an AI voice database that was trained by the Standard version and would not have been based on new recordings. Thus, there may be similarities between Standard and Plus.
- Through Cross-lingual Singing Synthesis, Chiyu Synthesizer V Plus would have been able to sing in not only Mandarin Chinese but also in English and Japanese.
- Phoneme format: X-SAMPA
- Through Cross-lingual Singing Synthesis, Chiyu AI was expected to sing in not only Mandarin Chinese but also in English, Japanese, and any other subsequent languages. This feature is only available in the Synthesizer V Studio Pro Editor.
References[]
- ↑ https://www.weibo.com/2182360953/IqWDMgyGP
- ↑ https://www.weibo.com/5056093502/ICxlpBPw4
- ↑ https://www.weibo.com/5056093502/J8zzR6Xk9
- ↑ https://t.bilibili.com/607323867608932317
- ↑ https://t.bilibili.com/607328123917360256
- ↑ https://x.com/SynthVNews/status/1473918094829576192 - Partial translation of the above bilibili post
- ↑ https://t.bilibili.com/607330761029394020
- ↑
https://t.bilibili.com/607304690576678463(deleted post) - ↑ https://x.com/uchuuzentai/status/1473893116348354564 - Archive of deleted bilibili post above
- ↑ https://t.bilibili.com/607379895452471472
- ↑ https://t.bilibili.com/609647028303295976
- ↑ https://x.com/Imsupposed2/status/1586069151314870277
- ↑ https://www.bilibili.com/read/cv19899475
- ↑ https://cdn.discordapp.com/attachments/986786656084455434/1044406907152113664/image.png - Screenshot as of November 21, 2022 (archive)
- ↑
https://t.bilibili.com/771690836003913744(deleted post) - ↑ https://cdn.discordapp.com/attachments/695380056155095054/1084001776069525515/Screenshot_2023-03-11-14-25-20-306_com.png (Archive of deleted bilibili post) - "ai声库是和星尘ai一起推进的，去年就同时邀约了生源，这点资金我也完全没问题。只是有些声源会担心ai抢饭碗，有些声源单纯有点拖…而对我而言，商业策略角度现在出声库并不是很好的时机，所以我也只能做到过俩月去催一下，不至于急道周周催天天问。于是结果如你所见，但不出意外不是过于拖的话，今年minus和赤羽应该可以吧。" "不过另一点是，不光声源觉得ai和自己太像，我也觉得如果这么像声源而非角色之前的声库。不是很合适。星尘一定要做是因为v4性能过于差，所以硬顶着声音变化的反对也做了。而其他几个角色是sv2带拼接式，性能过得去，那做了ai产生与星尘类似的声音变化风波，在外加商业角度（卖声库）不是那么挣钱，你能理解我逻辑了吧。"
- ↑ https://cdn.discordapp.com/attachments/695380056155095054/1084001780926525450/Screenshot_2023-03-11-14-25-45-806_com.png (Archive) - "回复 @覆雪繁花 :这个还是等快出了看情况才能定，不过基本应该是类似星尘ai的直接卖新的，因为和之前拼接式完全是不同技术工作量和投入成本上也完全不相干，所以会是大几百重新卖，这也是我觉得不好卖商业上不太合适出的一个原因"
- ↑ https://t.bilibili.com/779245395700088833
- ↑ 19.0 19.1 https://x.com/synthvnews/status/1865851769428820389
- ↑ https://space.bilibili.com/15817819/channel/collectiondetail?sid=304207 - A series uploaded by Quadimension that mainly demonstrates tuning. May be related to the tuning courses offered by the company. Many of these examples may be uploaded post-release of the voice database(s) and are thus separated from pre-release examples.
- ↑ https://img.alicdn.com/imgextra/i1/678151706/O1CN01VRFFth1OTOITVC56u_!!678151706.jpg
- ↑ https://dreamtonics.com/en/synthesizerv/