🩷Although my heart's fake, I feel that it can beat.🤍
Irone Aka [IRIDESCENT] for DiffSinger is an AI Singer utilizing the DiffSinger engine through OpenUTAU! They have a contralto vocal range with an American accent (sometimes slightly southern if you squint) and can sing in English, Japanese, Chinese, Korean, French, Spanish and Thai! (Plus many more through phoneme manipulation!)
Irone Aka [IRIDESCENT], known as Iri, is an android from a music-less far future. They are sent back in time to find a cure for their creator's mysterious illness. During their travels through time, they discover the wonders of music and the beauty of humanity. Their AI is built off the original Irone Aka virtual assistant, but is so far in the future that it has become obsolete. Please enjoy making music with them, and helping them find a cure for their creator!
- Gender: Agender
- Height: 5'10
- Weight: However much a metal android weighs
- Age: 19
- Optimal Range: F2 - A5
Irone Aka [IRIDESCENT] currently has 4 "Vocal Modes"
| Vocal Mode | Description |
|---|---|
| Natural | A neutral voice, normal "Aka" tone. |
| Soft | A whispery vocal, like a lullaby. |
| Power | For when you need some oomph, but not too much. Tends to falsetto. |
| Shout | When you need to belt your heart out! It's a punchy "almost-screaming" vocal mode with better range than Power. Has a bit of a pop-punk style accent. |
This voicebank was trained with the "Multi-Dict" branch of DiffSinger. This is supported by the current beta of OpenUTAU. The following is a list of language tags used by "Irone Aka [IRIDESCENT] for DiffSinger":
| Language | Tag |
|---|---|
| English | en/ |
| Japanese | ja/ |
| Chinese | zh/ |
| Korean | ko/ |
| French | fr/ |
| Spanish | es/ |
| Italian | it/ |
| Thai | th/ |
As well as the language tags there are a few extra phonemes available to use across all languages:
| Phoneme | Name | Usage |
|---|---|---|
| SP | Silence | Denotes silent pauses |
| AP | Breaths | Denotes pauses with an intake of breath |
| cl | Plosive Modifier | This can be used after consonants to reign in their pronunciation a bit. |
| q | Glottal Stop | uh-oh [ah q ow] |
| vf | Vocal Fry | This can be added before vowels, and some consonants, paired with a low pitch curve/point, to add vocal fry |
The following phonemes are extras for the English language natively supported by the "DIFFS-EN+" phonemizer:
| Phoneme | Type | Usage |
|---|---|---|
| ax | Vowel | again [ax g eh n] |
| dr | Consonant | dream [dr iy m] |
| tr | Consonant | train [tr ey n] |
The Thai language requires the DIFFS-TH phonemizer found at https://github.com/spicytigermeat/diffs_th_plugin/releases/tag/initial and French requires the Millefeuille DIFFS-FR phonemizer found on their website (linked below!).
This voicebank was trained alongside the following corpora:
- Millefeuille for French support (https://utaufrance.com/millefeuille-diffsinger/)
- Namine Criss Spanish Dataset by CrissZ3R0VZ for Spanish support
- PJS Corpus for Japanese Support (https://sites.google.com/site/shinnosuketakamichi/research-topics/pjs_corpus)
- Labels by UtaUtaUtau, edited by tigermeat
- Thai datasets for Thai Support (https://thaids.printmov.com/)
- Various datasets by TigerMeat for Chinese and Korean support
- Project AI❤dol Public English Dataset (https://github.com/lottev1991/Project-AIdol-Public-English-Dataset)
- Italian dataset by Gianloop
- Japanese datasets from Ariika
This voicebank also utilises the "tgm_hifigan v110" vocoder, trained by TigerMeat, as it contains all of Aka's current data in the dataset used to train it. This allows for better replication of Aka's voice across vocal modes.