This article is about the Synthesizer V Studio 2 software known as a voice database. If you are looking for the Synthesizer V concepts then click here.
History[]
2022-2023[]
According to Kanru Hua, the idea for building a voice database dedicated for choir vocals existed as early as 2022. Compared to other modern sample libraries on the market for choir voices, Kanru noted that traditionally these sample libraries were built with multiple pre-recorded vowels and syllables, allowing limited control of the lyrics or melody. After observation of users in 2022 building choirs with Synthesizer V Studio vocals by layering multiple solo voices, that of which are traditionally recorded in various different rooms and environments, with materials and microphones of differing qualities with different setups, and on different tracks, needing additional tweaking to achieve richer textures, Kanru noted that it could be difficult to get these vocals to sing together coherently as if they were singing in the same room. As a result, Kanru and the Dreamtonics Co., Ltd. team became curious on how close they could achieve a "real choir" with AI emulation.[1]
Kanru Hua and Dreamtonics Co., Ltd. would later collaborate with Junko Hirose, the official Dreamtonics operations manager (being in charge of logistics), who sourced and coordinated with choir singers for the project. Junko, who was familiar with watching televised choir performances such as those of Beethoven's Symphony No. 9, expected that this project would only be the recording of four lead singers: a soprano, an alto, a tenor, and a bass. She was later corrected to be recording four people per choir part: four sopranos, four altos, four tenors, and four basses altogether, for a total of 16 singers in total, who thought it would be a difficult undertaking to coordinate the schedules of. Due to a chance connection with a local Tokyo-based professional choir group, Dreamtonics began experimenting with recording the choir voice, placing a particular focus on the singers' spacing and the positioning of the microphones between each singer.[1]
Yongxin Ye, the official Dreamtonics sound engineer and data processor, noted that there were many questions such as wondering if there was a way to make individually-recorded solo voices sound close to a choir recording. An idea he suggested was placing a microphone in front of each singer of the choir in an attempt to record them as a solo voice. This idea was tested by verifying through a recording experiment and comparisons in-office. Many microphones with different directivity patterns were tested such as condenser microphones, dynamic microphones and "shotgun" microphones in an effort to remove microphone bleeding. Should the test be successful, it was hoped that they could record the solo voices and preserve the overall color of the choir. This experiment resulted in a hard-to-achieve result that was very sensitive, especially so when attempting to reproduce in a real studio. It was noted that there would be one or two singers at times who either were too far from the microphone or were singing too loud. For a comparison, Yongxin led a second experiment who invited the same group of singers to record the same song in a solo recording booth which involved essentially taking apart the choir and putting it back together, attempting to see if the results would still sound like a real choir.[1]
With the expectation of a flawless performance of each choir member, Kanru noted that after listening individually, the individuals sounded "drunk" in a sense that the singers' individual pitch and timing sounded off; however, when reassembled, they had a certain "magic" that was recognizable as a choir. Kanru noted that "baking" the same settings as those of the solo-recorded voice databases, one couldn't do "more" with the choir such as adding additional singers, being unable to tweak each one. Kanru had a goal of reaching the same level of detail as the solo-recorded voice databases that could be able to make use of the Synthesizer V Studio editor's various parameters and be tweaked. The model developed for the project was noted to be a small, but notably smart model that is trained on a real choir using multi-microphone recordings, the model being able to recognize the different placements with each singer singing slightly different in regards to the timing and pitch differences.[1]
After establishing a recording procedure in place, Junko stated that there was a need to apply it to not only the Japanese choir, but also to the English and Mandarin Chinese choirs. A clear theoretical direction was made to maintain quality, which was notably a difficult task across each session. The microphone positions, the singers' positions in the room, and the distance between each singer and their microphones needed to be closely monitored. During the first large-scale project with the Japanese choir recording, Yongxin and a fellow colleague were able to supervise the on-site sessions, then standardized the workflow for the English and Mandarin Chinese recording sessions. Yongxin found it challenging to find a way to supervise the remote sessions, such as reviewing the live renditions from respective local sound engineers, while providing feedback. Another challenge they needed to overcome was that the singers were not used to a tightly-controlled recording setup. Post-processing for noise reduction and breathing controls after each session were performed. It was noted that it took over one year to fully record all parts of the English and Chinese choirs. Various models were then trained finding many problems with the annotations, noises, or a singer that was singing too loudly. The project required many iterations to achieve a consistent texture that could sound like the real choir. Yongxin was notably very happy with how the three choir voices sound.[1]
2026[]
On January 15, 2026, Dreamtonics Co., Ltd. announced and simultaneously released the project as a new series of vocals, the Choir Voice Collections. Among them, three choir products –the first one in English, the second one in Mandarin Chinese, and the third one in Japanese– were released all together on all Dreamtonics stores (as well as on AH-Software Co. Ltd.'s AHS Store), either as separate collections (each one comprising of 16 vocalists each) or as a bundle encompassing all three collections, totaling 48 individual vocals. There were no prior public announcements. The collections were designed to take advantage of two new features introduced in the Synthesizer V Studio 2 Pro version 2.2.0 update, namely "Unison" and "Room Sound Simulation", that made it easier to create choruses in music production.[2]
Choir Voices 3, featuring Soprano, Alto, Tenor, and Bass voices specialized for singing "folk, hymnal, and ceremonial styles," was listed on the Dreamtonics International store as "Choir Voices – Collection 3",[2][3] on the Dreamtonics Taobao store as "Folk Choral Voices - Ling Guang Choir" (民谣合唱歌声 聆光合唱团; Mínyáo Héchàng Gēshēng - Líng Guāng Héchàng Tuán),[4] and on the AHS Store as "Ensemble Series vol.3" (アンサンブルシリーズ vol.3).[5] The Choir Voices bundle encompassing all three collections was listed on Dreamtonics International as "Choir Voices – Complete Bundle",[3] on Taobao as "Song Collection Set - 3-Choir Edition" (歌声合辑套装 ③合唱团版; Gēshēng Héjí Tàozhuāng - 3 Héchàng Tuán Bǎn),[4] and on AHS as "Ensemble Series vol.1~3 Set" (アンサンブルシリーズ vol.1～3セット).[5]
The voice databases were released at launch as version 202, along with seven-day trials.
Voice Database Information[]
Demonstrations[]
Some samples can be listened to on Dreamtonics's website.
| Demonstrations | |
|---|---|
|
| |
| 遥かなる追憶の彼方 (Harukanaru Tsuioku no Kanata) |  YouTube  bilibili
|
Voice Databases[]
Choir Voices 3 is an ensemble of 16 Japanese singers – 4 sopranos, 4 altos, 4 baritones (called "bass") and 4 tenors. As AI vocals, this enables them to sound realistic and human-like. Each singer can be managed individually by using the Unison panel.
It is to be noted that due to the parameters required to use properly these singers, Choir Voices 3 require Synthesizer V Studio 2 version 2.2.0 or more (including the betas of 2.2.0).- Choir Voices 3 are able 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, they can use AI Retakes to create multiple "takes" of a sung section to find an ideal vocal without the need for "laborious tuning". This feature is only available in the Synthesizer V Studio 2 Pro Editor.
- Each singer can be randomized in the Unison panel. This enables to change the vocal texture as well as the singing style of the vocal.
- It is possible to create a chorus up to 16 singers at the same time on a same track, all under the same vocal type (bass, tenor, soprano or alto).
- Phonetic format: Romaji
- Through Cross-lingual Singing Synthesis, Choir Voices 3 can sing in not only Japanese but also in English, Mandarin Chinese, Cantonese Chinese, Spanish, and Korean. This feature is only available in the Synthesizer V Studio 2 Pro Editor.
Choir Voices 3 comes with three variations within the Vocal Mode function:
- "Bright"
- "Rounded"
- "Smooth"
- OS: Windows 11/10 (64-bit) / Mac OS X 10.13 or newer.[6]
- AH-Software Co. Ltd.'s AHS Store recommends instead macOS 11 or later for the latter.[7]
- CPU: Intel Core i5 or newer / equivalent AMD Ryzen CPUs / Apple Silicon M1 or newer.[6]
- AHS Store recommends at least 4th generation for Intel Core i5 CPUs.[7]
- Memory: 4GB or more.[6][7]
- HDD: 1GB or more.[6][7]
- This is the space required when one voice database is installed.
- Space is required according to the number of voice databases to be installed.
- Others: audio device, internet connection required, email address able to send and receive messages required. (You need to create an account at Dreamtonics to use the product.)[7]
- Please check the product page for the latest operating environment. Even if you do not have a DVD-ROM drive, you can install the software via the Internet. Even if the software meets the requirements for operation, we do not guarantee that it will work on all computers.
- Dreamtonics Co., Ltd.'s End User License Agreement - Japanese
- Dreamtonics Co., Ltd.'s End User License Agreement - English
- Dreamtonics Co., Ltd.'s End User License Agreement - Simplified Chinese
Please note singer letters are determined by the brightness level of their head in the Unison panel. Singer a is the one loaded by default, represented by a head with 100% brightness level. Singer b has a 25% brightness level, singer c a 50% level and singer d a 75% level. These letters are NOT official and are used to determine which singer in the Unison panel has been chosen.
Choir Voices 3 Solfeges:
- Alto (no vocal mode)
- (a)
- (b)
- (c)
- (d)
- (a)
- Bass (no vocal mode)
- (a)
- (b)
- (c)
- (d)
- (a)
- Soprano (no vocal mode)
- (a)
- (b)
- (c)
- (d)
- (a)
- Tenor (no vocal mode)
- (a)
- (b)
- (c)
- (d)
- (a)
- Alto (Bright - 150%)
- (a)
- (b)
- (c)
- (d)
- (a)
- Bass (Bright - 150%)
- (a)
- (b)
- (c)
- (d)
- (a)
- Soprano (Bright - 150%)
- (a)
- (b)
- (c)
- (d)
- (a)
- Tenor (Bright - 150%)
- (a)
- (b)
- (c)
- (d)
- (a)
- Alto (Rounded - 150%)
- (a)
- (b)
- (c)
- (d)
- (a)
- Bass (Rounded - 150%)
- (a)
- (b)
- (c)
- (d)
- (a)
- Soprano (Rounded - 150%)
- (a)
- (b)
- (c)
- (d)
- (a)
- Tenor (Rounded - 150%)
- (a)
- (b)
- (c)
- (d)
- (a)
- Alto (Smooth - 150%)
- (a)
- (b)
- (c)
- (d)
- (a)
- Bass (Smooth - 150%)
- (a)
- (b)
- (c)
- (d)
- (a)
- Soprano (Smooth - 150%)
- (a)
- (b)
- (c)
- (d)
- (a)
- Tenor (Smooth - 150%)
- (a)
- (b)
- (c)
- (d)
- (a)
References[]
- ↑ 1.0 1.1 1.2 1.3 1.4 https://www.youtube.com/watch?v=24tiChS6DOo
- ↑ 2.0 2.1 https://dreamtonics.com/choir-voice-collections/
- ↑ 3.0 3.1 https://store.dreamtonics.com/product/choir-voices-collection/
- ↑ 4.0 4.1 https://item.taobao.com/item.htm?id=1014345156016
- ↑ 5.0 5.1 https://www.ah-soft.com/synth-v/dreamtonics2/chorus.html
- ↑ 6.0 6.1 6.2 6.3 https://store.dreamtonics.com/product/synthesizer-v-studio-2-pro/ (February 21, 2025 archive)
- ↑ 7.0 7.1 7.2 7.3 7.4 https://www.ah-soft.com/synth-v/sp/#spec (February 18, 2025 archive)