SynthV Wiki
Advertisement
Hanakuma Chifuyu
SynthV Studio
CeVIO AI
VoiSona Talk
VOICEPEAK
|
Originals
|
Gallery

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[]

Originally, Natsuki Karin and Hanakuma Chifuyu's voice synthesizer production announcement was meant to be announced during Koharu Rikka's first live concert, "Koharu Rikka 1st LIVE! First Day of School!", in Fall 2020. However, because the concert was postponed due to the COVID-19 pandemic, the pair's production was unannounced.[1]

2021[]

On May 20, 2021, a silhouette was shown on TOKYO6 ENTERTAINMENT's twitter account. Details were to be released on May 22.[2] This was later revealed to be Hanakuma Chifuyu, a character introduced through a campaign titled "Koharu Rikka x Otaru Collaboration (小春六花×小樽コラボ)". She was confirmed to be a student enrolled at the fictional Otaru Shiokaze High School and was in the same music club as Koharu Rikka.[3][4] Details about Chifuyu as a character were released and she was confirmed to be voiced by Kaya Okuno for a promotional video.[5][6] She was illustrated by Teshima Nari and the logo was created by Hiroaki Watanabe.[7] It was noted in a TOKYO6 ENTERTAINMENT livestream that there were plans to create a voice library for Chifuyu and Natsuki Karin, which were confirmed in June to be Synthesizer V AI and CeVIO AI after TOKYO6 ENTERTAINMENT adjusted their website.[8]

On August 8, due to some concerns regarding business activities related to Synthesizer V, Dreamtonics Co., Ltd. noted that Chifuyu was on the list of "yet-to-be-published" voices and initiatives.[9] For many overseas fans, this was the first time they became aware of these voice databases as there was no official announcement via social media nor newsletter of the productions and website adjustments. Due to language barriers, the livestream featuring the announcement was also overlooked.

On October 3, it was announced that Chifuyu would be the third character in the lineup, meaning that Karin was to be prioritized over her. A crowdfunding campaign for Chifuyu would be launched around June 2022 with details to release in the spring.[10]

On October 26, it was confirmed via livestream that recordings for a test Chifuyu bank had begun and would continue onto the next day. Samples were expected to be provided in the spring.[11]

2022[]

On January 4, 2022, TOKYO6 confirmed that crowdfunding for her Synthesizer V AI voice database and CeVIO AI Talk Song library would start this year and that their production was planned to be finished before the year's end.[12]

On February 16, Chifuyu's official Twitter feed reconfirmed the crowdfund to take place in June, with aims to get both of her CeVIO AI and SynthV AI libraries released in October.[13] During the AHS live broadcast the next day (where TOKYO6 acknowledged their cooperation in the development of VOICEPEAK, AH-Software's new text-to-speech software based on Dreamtonics' Syllaflow AI speech-synthesis engine), her October release month was confirmed at least for her CeVIO AI Talk voice.[14]

On March 28, it was announced that Chifuyu's CeVIO AI and Synthesizer V AI production crowdfund was scheduled to launch on June 26. Details were to come in mid-May.[15]

On April 22, Chifuyu's official Twitter announced that "she" would be performing at "Koharu Rikka 1st LIVE! First Day of School!".[16] At the concert, Kaya Okuno performed "ヒメタルネ" ("Himetaru.NE")[17], a song written by Itsuki Minato, Takapira's keyboardist, that was intended to be a future demo for Chifuyu.[18][19] On May 2, TOKYO6 tweeted that a live broadcast was scheduled for May 20 at 19:30 JST with appearances from Yoshino Aoyama, Miyu Takagi, Kaya Okuno, and Akasako. Discussions about Rikka's birthday festivities, "Koharu Rikka 1st LIVE! First Day of School!", Karin's release commemoration, and Chifuyu's crowdfund outline for June 26 were to be expected.[20] Chifuyu's CeVIO AI Talk Voice and Synthesizer V AI voice database were confirmed to be showcased at the live broadcast.[21]

On May 20, during TOKYO6's livestream, Chifuyu's crowdfunding was announced to begin on June 26 at 9 PM JST during another broadcast and would be hosted on ubgoe. It was scheduled to end on July 26 at 23:59 JST. The first goal, at 6 million yen, included the confirmation of Chifuyu's CeVIO AI Talk Voice and Synthesizer V AI voice database. The second goal, at 8 million yen, included 800 PitaGoe samples. The third goal, at 10 million yen, included a 3D model and an additional demo. Afterwards, for each time 3 million yen was raised, another demo and standing visuals would be added.[22] The crowdfund page was made available on June 20 and was still scheduled to begin on June 26, with an end date of July 26. It was noted that backers would be receiving Chifuyu around nine days prior to her release and it was asked that backers refrain from posting any works using Chifuyu until the day before her release at 20:00 JST.[23] The main crowdfund goal was reached within minutes of launch, and then the final stretch goal was reached after an hour.[24]

On a livestream hosted on June 26, TOKYO6 ENTERTAINMENT mentioned that five Vocal Mode variations were being implemented for Chifuyu and hoped to have them ready by the time she was released. These were confirmed to be called "Kawaii", "Soft", "Rock", "Adult", and "Piano_Ballade".[25][26][27]

On September 25, listings for Chifuyu's physical box were made available for pre-order on Amazon Japan with no prior announcement. According to the listing, she was planned to release on October 20.[28] The listing was later taken down, with TOKYO6 ENTERTAINMENT tweeting that they were making final adjustments with retailers to release Chifuyu's products in mid-October, marking the release date as undecided for the time being.[29] A follow-up tweet was posted to remind crowdfund supporters to adjust their addresses if necessary.[30] On September 30, AH-Software announced that Chifuyu's CeVIO AI and Synthesizer V AI voices would be released on October 20 as physical and digital copies. She was also eligible for the Synthesizer V Studio Pro Starter Pack.[31] According to her AH-Software page, her Lite version would be released in November and she did not support the AI Retakes function at the time.[32][33] Shipment for crowdfund backers was scheduled for October 7.[34]

Due to the additional funds gained from the campaign in June, on September 29, TOKYO6 ENTERTAINMENT announced that there would be four additional standing art (illustrated by MtU, Karei, Niijima Akiichi, and Yume no Owari) and a total of six demos for Chifuyu, which included the previously shown "Himetaru.NE". Half of the demo songs were to be launched on October 19, the day before her release, while the other half were scheduled for February 6, 2023.[35] Some of the demos were confirmed to be: a song produced by Kazawara, a song produced by Minato Itsuki, the full version of "Himetaru.NE", a trio song, and more.[36] Crowdfund rewards, including the voice database, would be shipped on October 7.[37] On October 19, Chifuyu's demos were uploaded and crowdfund backers were invited to post their works with the "#花隈千冬" tag.[38] AH-Software also announced that shipment for Chifuyu's physical CeVIO and Synthesizer V packages was scheduled for the day.[39] Chifuyu released on October 20 as expected. The Lite voice database was released later on November 2, though was momentarily suspended due to problems with functions before being reinstated for download.[40][41][42]

On November 14, after the release of Chifuyu's v102b1 installer, TOKYO6 ENTERTAINMENT released a survey for its feedback.[43] On November 22, the company noted that the beta version would be used as the official update after receiving mostly positive feedback.[44]

Updates[]

Hanakuma Chifuyu AI v102b1 - November 10, 2022[45] 
A free update for Hanakuma Chifuyu was released following the Synthesizer V Studio 1.8.0b1 update. This enables her to use the new Diffusion Probabilistic Models (DPM) feature.
Hanakuma Chifuyu AI v102 - November 24, 2022[46][47] 
Final version update compatible for Synthesizer V Studio 1.8.0 which enables Hanakuma Chifuyu to use the new Diffusion Probabilistic Models (DPM) feature if the beta version was not installed previously.
Hanakuma Chifuyu AI v103b1 - June 21, 2023[48][49] 
A free update for Hanakuma Chifuyu was released following the Synthesizer V Studio 1.9.0 update. This implemented the new rap feature available for the English and Mandarin languages. Cantonese Chinese was added to the Cross-lingual Singing Synthesis feature as well as general improvements to the overall function, resulting in more accurate pronunciation that sounds closer to a native speaker when singing foreign words.
Hanakuma Chifuyu AI v103 - July 6, 2023[50][51] 
Final version update compatible for Synthesizer V Studio 1.9.0 which enables Hanakuma Chifuyu to use the rap feature (English and Mandarin Chinese only) and allows for Cantonese Chinese Cross-lingual Singing Synthesis if the beta versions were not installed previously.
Hanakuma Chifuyu AI v104b1 - August 2, 2023[52][53][54][55] 
A free update for Hanakuma Chifuyu was released following the Synthesizer V Studio 1.10.0b1 update. This enhanced pitch generation by utilizing Reinforcement Learning from Human Feedback (RLHF), translating to fewer off-key errors, making better use of vibrato, and creating more expressive transitions between notes. This also allowed for the use of the new "Enhancement" slider in the AI Retakes panel, which controlled the level of changes introduced by RLHF. For Synthesizer V Studio Pro users, feedback submission was enabled, allowing users to consensually share their preferred takes with only Dreamtonics for research and development purposes, where models trained with this feedback would be deployed in future voice database updates. Users may also opt out of this feature if undesired.
Hanakuma Chifuyu AI v104b2 - September 28, 2023[56][57][58] 
A free update for Hanakuma Chifuyu was released following the Synthesizer V Studio 1.10.0b2 update, which was the second beta version following version 1.10.0b1 that introduced enhancements to voice databases and integrated the latest user feedback via Reinforcement Learning from Human Feedback (RLHF). Regarding the editor, numerous reported issues were addressed for a more fluid experience. A new generation of pitch models were trained using the extensive feedback from users over the past month, and the new models offered greater accuracy in voice timbre reproduction, as well as a reduction in mispronunciations. Dreamtonics also focused on issues from the previous beta release that made Vocal Modes seem less distinct on certain vocals, ensuring clearer differentiation in the 1.10.0b2 update. The latest update also introduced a new feature in the Note Properties Panel that allowed for even division of a note into multiple syllables, proving to be a useful tool for songs with faster tempos.
Hanakuma Chifuyu AI v106b1 - November 24, 2023[59][60][61] 
A free update for Hanakuma Chifuyu was released following the Synthesizer V Studio 1.11.0b1 update. This implemented support for the Japanese language in the rap feature. In addition, it disables Reinforcement Learning from Human Feedback (RLHF) enhancement for pitch generation by default, as requested by TOKYO6 ENTERTAINMENT.[62][63]
Hanakuma Chifuyu AI v106 - April 4, 2024[64][65][66] 
Final version update compatible for Synthesizer V Studio 1.11.0 which disables RLHF by default and enables Hanakuma Chifuyu to rap in Japanese, if the beta version was not installed previously. In addition, this update implements support for singing in the Spanish language in the Cross-lingual Singing Synthesis feature.

Voice Database Information[]

Demonstrations[]

Demonstrations

ヒメタルネ (Himetaru.NE) ft. Koharu Rikka - Short Version Favicon-nnNicoNico Favicon-ytYouTube Favicon-urlTwitter
アレカコニア (Arekakonia) - Short Version Favicon-ytYouTube
Dancing in the binary - Short Acapella Version; English Cross-lingual Singing Synthesis Favicon-ytYouTube
远方有星辰 (Yuǎnfāng Yǒu Xīngchén) - Short Acapella Version; Mandarin Chinese Cross-lingual Singing Synthesis Favicon-ytYouTube
あいにいく (Ai ni Iku) Favicon-nnNicoNico Favicon-ytYouTube Favicon-bbbilibili
ツギハギビート (Tsugihagi Beat) Favicon-nnNicoNico Favicon-ytYouTube Favicon-bbbilibili
ヒメタルネ (Himetaru.NE) ft. Koharu Rikka - Full Version Favicon-nnNicoNico Favicon-ytYouTube Favicon-bbbilibili
季節の交差点 (Kisetsu no Kousaten) ft. Koharu Rikka & Natsuki Karin Favicon-nnNicoNico Favicon-ytYouTube Favicon-bbbilibili Favicon-urlpiapro
SEA you!! ft. Koharu Rikka & Natsuki Karin Favicon-nnNicoNico Favicon-ytYouTube Favicon-bbbilibili
モンスターズ (Monsters) Favicon-nnNicoNico Favicon-ytYouTube Favicon-bbbilibili
Winter Parade Favicon-nnNicoNico Favicon-ytYouTube Favicon-bbbilibili
先生あのね (Sensei Ano ne) Favicon-nnNicoNico Favicon-ytYouTube

Voice Databases[]

Product Information
Optimum
  Range: F3~G5

Versions
  Lite Version/FLT Available?: Yes (Lite)  Starter Pack Available?: Yes  Web Version Available?: No

Package details as noted:
Hanakuma Chifuyu is a youthful feminine Japanese voice database with a rich bass tone and exquisite breathiness. Her voice was noted to be suitable for all genres and is based on the singing style of Kaya Okuno.[67] As expected of an AI voice database, this enables her to sing with a more human-like and realistic voice.

Vocal traits as noted:
  • It was noted that Chifuyu's voice type, that being a blasé-like voice with a high breath component, was difficult to reproduce through conventional voice synthesis software.[1]
  • The breathiness gives the voice a sweet feel.[1]
  • She has a strong lower register that doesn't break when she sings ballads.[1]
  • The slight lingering and sadness in her voice when transitioning from low to high notes was considered appealing.[1]
  • Initially, upon the publishing of her product page at the time, Chifuyu was not yet compatible with AI retakes, a feature via High Dynamics Voice Model (HDVM). This feature would have allowed Chifuyu to sing more realistically through changes in volume and breath, further enriching singing details, making each singing unique, and providing creators with more options. Through this mode, she could have used AI Retakes to create multiple "takes" of a sung section to find an ideal vocal without the need for "laborious tuning". HDVM was later migrated to Diffusion Probabilistic Models (DPM) per the 1.8.0b1 update on November 10,[68] which Chifuyu was made compatible with via her v102b1. This feature is only available in the Synthesizer V Studio Pro Editor.
    • This does not apply to the Lite version.

Phonetic notes as noted:
  • Phoneme format: Romaji
  • Through Cross-lingual Singing Synthesis, Chifuyu can sing in not only Japanese but also in English and Mandarin Chinese. Since the release of her v103b1 update on June 21, 2023, she can also sing in Cantonese Chinese with this feature; since her v106 update on April 4, 2024, she can also sing in Spanish with this feature. Cross-lingual Singing Synthesis is only available in the Synthesizer V Studio Pro Editor.
    • This does not apply to the Lite version, thus restricting the voice database to Japanese only.
      • An issue with the initial release of the Lite voice database allowed it to work with the Cross-lingual Singing Synthesis function. This had since been replaced with a new version that prevented it from using the function.[41]
  • Downloadable Pitagoe voice files are available for those with Chifuyu's CeVIO AI Talk Voice or Synthesizer V AI voice database after user registration on AH-Software's website. There are over 800 audio files to use for a wide variety of genres.[69]

Vocal Modes as noted (Synthesizer V Studio Pro Editor only):
  • The paid version of Hanakuma Chifuyu comes with five variations within the Vocal Mode function:
    • "Kawaii"
    • "Soft"
    • "Rock"
    • "Adult"
    • "Piano_Ballade"
  • An issue with the initial release of the Lite voice database allowed it to work with the Vocal Mode function. This had since been replaced with a new version that prevented it from using the function.[41]

System requirements as noted:

Hardware Requirements:

  • CPU:
    • Intel Core i3 or higher recommended
  • RAM: 2GB or higher
  • Disk Space: 1 GB or higher (for installing one voice database)
    • The amount of space required depends on the number of voice databases installed.
  • Display: 1280 x 800 or higher

Software Requirements:

  • Operating System:
    • Windows 10/8.1
    • Mac OS X: 10.11 or higher
    • Linux Ubuntu 20.4 or higher

Other:

  • DVD-ROM Drive (physical edition only)
  • Sound card
  • Internet connection is necessary for activation and obtaining updaters.

License agreements as noted:
  • TOKYO6 ENTERTAINMENT's Character Usage Guidelines - Japanese
  • AH-Software's End User License Agreement - Japanese
  • AH-Software's End User License Agreement - English
  • Regarding the Lite version:
    • AH-Software does not accept technical inquiries regarding free software nor do they guarantee or support the operation of all free software.
    • AH-Software is not liable for any damages (including loss of data) caused by using free software. Users are cautioned to use the free software at their own risk.
    • The program may not be transferred, sold, lent, distributed or rented without AH-Software's permission, regardless of whether it is paid or free of charge.
    • Reverse engineering, including modification, decompilation, and disassembly, of their program for any purpose is prohibited.
    • Hanakuma Chifuyu 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.

Voice database sample

References[]

  1. 1.0 1.1 1.2 1.3 1.4 https://www.dtmstation.com/archives/60921.html
  2. https://twitter.com/tokyo6info/status/1395588323251425283
  3. https://twitter.com/tokyo6info/status/1396090971088211973
  4. https://twitter.com/tokyo6info/status/1396092229597470723
  5. https://twitter.com/tokyo6info/status/1396092229253484545
  6. https://note.com/tokyo6/n/nc5e32c933fa1
  7. https://twitter.com/tokyo6info/status/1396092480836100103
  8. https://www.youtube.com/watch?v=M9wDSETvuzY
  9. https://twitter.com/dreamtonics_en/status/1424533859752812545
  10. https://twitter.com/tokyo6info/status/1444648429918081026
  11. https://www.youtube.com/watch?v=xLcrmYPSpM0
  12. https://twitter.com/tokyo6info/status/1478171466361610241
  13. https://twitter.com/Chifuyu_Info/status/1493892011271008258
  14. https://twitter.com/tokyo6info/status/1494296032750219267
  15. https://twitter.com/tokyo6info/status/1508274508536643586
  16. https://twitter.com/Chifuyu_Info/status/1517760182486548480
  17. https://twitter.com/kaayaataaaan/status/1518219391048896512
  18. https://twitter.com/s_akasakov/status/1518268067771744256
  19. https://note.com/tokyo6/n/nf413e2ef33cd
  20. https://twitter.com/tokyo6info/status/1521058584918114305
  21. https://twitter.com/Chifuyu_Info/status/1526738345531183104
  22. https://www.youtube.com/watch?v=vvpaI_qIbS4
  23. https://twitter.com/tokyo6info/status/1538718530371301377
  24. https://twitter.com/tokyo6info/status/1541070939374624768
  25. https://www.youtube.com/watch?v=1EM0irCWP_Y
  26. https://cdn.discordapp.com/attachments/695380056155095054/990795122641891359/Screenshot_2022-06-26-18-09-48-902_com.google.android.youtube.jpg - Screenshot as of June 26, 2022
  27. https://twitter.com/endryu36012867/status/1578579368502493187
  28. https://amzn.asia/d/hySylHc
  29. https://twitter.com/tokyo6info/status/1574320537035161600
  30. https://twitter.com/tokyo6info/status/1574320538494783488
  31. https://www.ah-soft.com/press/cevio/20220929.html
  32. https://www.ah-soft.com/synth-v/chifuyu/index.html
  33. https://cdn.discordapp.com/attachments/695380056943493280/1024906113549877339/unknown.png - Screenshot as of September 30, 2022
  34. https://twitter.com/Chifuyu_Info/status/1575326844064129024
  35. https://twitter.com/tokyo6info/status/1575391532953190400
  36. https://twitter.com/tokyo6info/status/1575391532953190400
  37. https://twitter.com/tokyo6info/status/1578194465621434368
  38. https://twitter.com/Chifuyu_Info/status/1582237074543247360
  39. https://twitter.com/ahsoft/status/1582630849509347329
  40. https://twitter.com/ahsoft/status/1587716481793503232
  41. 41.0 41.1 41.2 https://twitter.com/ahsoft/status/1587755085915824129
  42. https://twitter.com/ahsoft/status/1587770232243773446
  43. https://twitter.com/tokyo6info/status/1591981747180163072
  44. https://twitter.com/tokyo6info/status/1594852826085494784
  45. https://dreamtonics.com/en/synthesizer-v-studio-1-8-0b1-update/
  46. https://dreamtonics.com/en/synthesizer-v-studio-1-8-0-final-update/
  47. https://twitter.com/ahsoft/status/1595674717200916481/
  48. https://dreamtonics.com/synthesizer-v-studio-1-9-0-final-update/
  49. https://twitter.com/dreamtonics_en/status/1671427830629154816
  50. https://www.bilibili.com/read/cv24826703/
  51. https://twitter.com/ahsoft/status/1676865674898206720
  52. https://twitter.com/dreamtonics_en/status/1686648078571655168
  53. https://t.bilibili.com/825177687051993121
  54. https://dreamtonics.com/synthesizer-v-studio-1-10-0b1-update-enhancing-pitch-generation-with-user-feedback/
  55. https://twitter.com/ahsoft/status/1686653272474738689
  56. https://twitter.com/dreamtonics_en/status/1707304150059639108
  57. https://dreamtonics.com/synthesizer-v-studio-1-10-0b2-update/
  58. https://twitter.com/ahsoft/status/1707304906535956719
  59. https://dreamtonics.com/synthesizer-v-studio-1-11-0b1-update/
  60. https://twitter.com/ahsoft/status/1727970395691856025
  61. https://twitter.com/tokyo6info/status/1727987685288206599
  62. https://note.com/tokyo6/n/n742c974716b9 - The main reasoning behind TOKYO6's request to disable RLHF by default (which prevented the release of finalized versions of Koharu Rikka AI v124, Natsuki Karin v110, and Hanakuma Chifuyu v106 and also resulted in the permanent withdrawal of their last respective beta versions) was that the then-current implementation of RLHF learned AI-retake data selected as favorable by various different users of Rikka AI, Karin, and Chifuyu and then reflected it across all installations of their voice databases, thus affecting their overall preference trend; TOKYO6 believed that, due to things like variations in preferred vibrato swing size from user to user, etc., some users would get undesirable results out of their vocals, and in internal beta test results, TOKYO6 noted that the vocals' vibrato tended to be exaggerated and deviating from their character image. TOKYO6 was also concerned about RLHF results automatically affecting the default singing style (with no Vocal Modes applied) to a large extent, resulting in the decision to postpone any further updates to their voice databases until they were given the possibility to disable RLHF by default in the next update.
  63. https://twitter.com/tokyo6info/status/1717458061949026415
  64. https://dreamtonics.com/synthesizer-v-studio-1-11-0-update/
  65. https://twitter.com/ahsoft/status/1775800933538320621/photo/4
  66. https://twitter.com/tokyo6info/status/1775804166952403019
  67. https://ubgoe.com/projects/200
  68. https://dreamtonics.com/en/synthesizer-v-studio-1-8-0-final-update/
  69. https://twitter.com/tokyo6info/status/1578194540783337472

Navigation[]

Advertisement