| Release Date | |
| Recording Languages | English & Japanese |
| Additional Languages | Spanish + Italian Trained with Gianloop’s dataset + Criss Namine’s Dataset French Trained with UTAU French Resources’ Petite Millefeuille Korean Trained with CSD (labels by heta-tan) Mandarin Chinese Trained with OpenCPop (labels by heta-tan) |
| Recording Range | D3 – C6 |
| Voice Colors | Main Sequence, White Dwarf, Nebula, Red Giant, Neutron, Protostar, Supernova, Black Hole |
Last updated: July 7th, 2025
Current voicebank version: V7
Page and voicebank are still WIP!
Thank you for your patience!
Stellar Evolution is the process in which a star changes over time. Starting from stellar nurseries such as Nebulae or molecular clouds, matter condenses to form a Protostar. Once the star starts to stably burn hydrogen into helium, it’s considered to be a Main Sequence star, where it will spend about 90% of its life cycle. As stars run out of hydrogen to convert, the core will begin to collapse, becoming a Red Giant. Depending on the star mass, it can take one of two routes: low mass stars will see their outer layers blow away, creating a planetary Nebula and the remain core becoming a dense White Dwarf. Larger mass stars can explode into a Supernova, collapsing into either an extremely dense Neutron star or a Black Hole.
Aphelia -stellar evolution- is an AI singing voice bank made using DiffSinger and singing data from their voice provider, heta-tan. -stellar evolution- offers the most realistic and versatile version of Aphelia. -stellar evolution- is officially able to sing in several languages including: English, Japanese, French, Italian, Spanish, Korean, and Mandarin. As well, they include 7 different voice colors to help them sing a wide range of genres and singing styles.
Change Log (Click to show)
V7
- Re-recorded large sections of voicebank (Mainly Main Sequence + Red Giant + Supernova) to reduce reverb
- Re-added Black Hole (whisper) voice color
- Added singing Japanese recordings to all voice color
- Added more growl recordings
- Added more data for range
- Trained using multi-dict configuration
- Added PJS Japanese database
- Added Criss Namine Spanish database
- Trained with Muon Lynxnet2
- Trained Aphelia_PC_hifigan to have in-house tone shift support
V6
- Trained with lynxnet +melody encoder
- Re-recorded large sections of Main Seqeuence and Red Giant to reduce reverb
- Re-recorded and added large sections of Supernova to increase quality
- Added new phoneme: gr (growl VERY EXPERIMENTAL!)
- Added more data for range
- Added Korean support using CSD
- Added Mandarin support using OpenCPop
- Edits to phoneme system to make it more intuitive
V5
- Full Dataset!
- Added new phonemes: cr (voice creak), MM (closed mouth humming), NN (open mouth humming) and LL (long L humming?)
- Began working on Portuguese replacement dictionary (HEAVY WIP!)
- Added DIFF JA support
- Updated vocoder (aphelia_nsfgan) to include full dataset
- Text tutorial is mostly drafted
V4
- Mostly all new data
- Added Spanish/Italian support thanks to Gianloop
- Added French support thanks to UTAU French Resources
- dsdict-en.yaml has been added to improve pronunciation and allow for easier use of additional phonemes
- Phonemes have been changed
- rr/r (JPN) have been merged and changed to [rx]
- h (JPN) has been changed to [hh] to merge with English data
- jh (JPN) has been changed to [jh] to merge with English data
- Custom vocoder (aphelia_nsfgan) has been included
- Vocal mode -moon- (Japanese) has been merged with -main sequence-
- Vocal mode -supernova- (Rock) has been added
Main Sequence is the default voice color for this voicebank and is the most similar to Aphelia’s VCCV English and their -capricornius- /-ophiuchus- banks. Main Sequence tends to have a dark, mature and masculine lower range that transitions to a youthful, sweet feminine higher range.
| Sample of Recorded Genres | Jazz, Pop-Rock, Soul, City Pop, Pop |
| Recording Range | D3 ~ A5 |
| Recording Amount | 1:53:09 |
White Dwarf is the soft voice color. This voice color is most similar to their -pisces- or -corona- ‘s whisper expression. It’s a very soft and airy voice with delicate pronunciation while retaining qualities of Main Sequence. This voice works well in tandem with Nebula.
| Sample of Recorded Genres | Acoustic, Bossa Nova, R&B, Easy Listening |
| Recording Range | E3 ~ E5 |
| Recording Amount | 0:52:57 |
V.5 Demonstration
Nebula is a mature, deep, soft voice that can take on an almost operatic approach in its higher register. This voice color is most similar to their -cygnus- or -corona-‘s dark expression. This voice is full of emotion and can sometimes take on a sultry quality. You can use this voice color together with Neutron as a soft counterpart.
Designed by Gardanana
| Sample of Recorded Genres | Ballads, Folk, Atmospheric, Jazz, Soul, Acoustic, Opera |
| Recording Range | F3 ~ F5 |
| Recording Amount | 0:51:09 |
V.5 Demonstration
Red Giant is considered the “default power” voice of this voicebank. This voice color is most similar to their -solar flare- base and light expressions. It is a confident and bold voice with clear pronunciation while still retaining qualities of Main Sequence. The higher range for this voice color is very bright and strong with a very cheerful tone. This voice is a more powerful version of Main Sequence and works well blended with Neutron.
| Sample of Recorded Genres | Disco, Pop, Funk |
| Recording Range | G3 ~ B5 |
| Recording Amount | 0:49:08 |
Neutron is a powerful compliment to Nebula. This voice color is most similar to -stratosphere- chest expression. It is a more mature, chest based vocal compared to Red Giant that can have looser and more exaggerated pronunciation. This vocal is more of an operatic type of vocal than it is a rock/typical power vocal.
Designed by liv/ricecrispty
| Sample of Recorded Genres | Symphonic, Musical Theater, Alternative/Indie, Folk-Rock, Opera |
| Recording Range | E3 ~ D5 |
| Recording Amount | 0:52:03 |
V4 Demonstration
Protostar is voice acted, childish and nasally voice color compared to the rest of the voicebank. This voice color is most similar to -corona- ‘s sweet expression or to Ame Wakana, another UTAU voiced by heta-tan. This expression is mainly for fun so it may not produce the best vocals for serious projects, but they might be useful in XSY as a pseudo-gender flag or for more cartoony vocals.
Designed by ztarplatinum
| Sample of Recorded Genres | Children’s Songs, Bubblegum pop |
| Recording Range | G3~F5 |
| Recording Amount | 0:29:02 |
V4 Demonstration
Supernova is a powerful and playful voice. This voice mode is most similar to the -solar flare- yell expression. Unlike Neutron, this voice color is less chest based, having a much better higher range for belting vocals. Unlike Red Giant, there’s more exaggerated and looser in its singing style and pronunciation. Sometimes, this vocal can have a gritty quality to them, especially in the higher range.
| Sample of Recorded Genres | Classic Rock, Vkei, Pop Punk, Hard Rock |
| Recording Range | G#3~C6 |
| Recording Amount | 0:38:58 |
Black hole is a whispering vocal. This is an experiment vocal to see if DiffSinger can handle very airy, nearly whispering vocals with the help of “trash” phonemes. With negative tension values, you can obtain 100% whispering vocals. This voice color is well suited for harmonies, ad-libs and “effect” vocals instead of primary vocals.
| Sample of Recorded Genres | Children’s Songs |
| Recording Range | E3~D5 |
| Recording Amount | 0:09:31 |
DOWNLOADS
If you are having issues with the mediafire download, I’ve also uploaded the voicebank to Google Drive as well!
Bonus Goodies
What’s Included
Aside from a DiffSinger voicebank that can sing in English, Japanese, French, Spanish, Italian, Korean and Mandarin, Aphelia’s -stellar evolution- package will include the following:
- 0_Extras
- exVoice Link
- Additional recordings (EN/JA): laughter, counting, adlibs, etc.
- ✓ Allowed: covers, creative projects
- X Prohibited: training AI models (RVC, soVITS, Diff-SVC, etc.)
- Resources
- Tutorial PDFs
- Phoneme charts
- Base dictionary files
- exVoice Link
- Vocal Color Archives
- Each voice includes:
- Full body art (HD + OpenUTAU-optimized)
- Icons
- Logos (DiffSinger / voice color / Stellar Evolution)
- Additional artwork
- Reference sheets + design notes
- Each voice includes:
Requests and feedback for new features can be sent through this Google Form
| Bank Type | Upload Date | Recording Range | Vocal Modes |
|---|---|---|---|
| Multilingual DiffSinger | July 7th, 2025 | D3-A5 | Main Sequence White Dwarf Nebula Red Giant Neutron Protostar Supernova Black Hole |
Deep Space UTAU 