illust by OtomeHime

KEIRO AI
for DiffSinger

✦ multilanguage: Russian and Japanese
✦ 2 voice colors: standard and power
✦ data size (w/o silence):
. standard: 68 minutes
. power: 23 minutes
✦ supports Russian (cyrillic) words and Japanese (hiragana) characters
✦ has 7 additional phonemes for good Japanese pronunciation
✦ supports vocal fry (vf) and glottal stop (cl)
✦ recommended range: F2 - A4

KEIRO AI
for Diff-SVC

✦ 44.1 kHz
✦ about 6 hours 20 minutes of data
✦ 216k steps
✦ trained on samples for utau, singing and speaking on English, Russian, Japanese and Chinese