illust by ポメポメ

Nishiren Gard (西蓮ガルド) is a vocal synth that can sing using Diffsinger for OpenUTAU. They have a contralto voice with a youthful and versatile tone that can easily be adjusted to fit a wide variety of music. Nishiren sings primarily in English and Japanese, but can also sing in a few additional languages.

Voicer/Manager: Gardanana
Official Site: https://gardanana.neocities.org/

DOWNLOAD LATEST (v2.0)
By downloading this voicebank you agree to their Terms of Service

For details on installing and using the voicebank, check out the Usage Guide

TECHNICAL INFORMATION

Range: Contralto (A2-A5)
Microphones: Samson Go, AT2020USB+, Neumann U87
Available Parameters: gender (GENC), velocity (VELC), tension (TENC), breathiness (BREC), voicing (VOIC), autopitch
Recorded Languages: English, Japanese
XLS: Mandarin Chinese, Korean, Spanish, Italian, French, Thai

Phonemizers:

[DIFF] used for both Kana and Pinyin inputs
[DIFF EN] used for English, Kana(JA), and Hangul(KO) inputs
[DIFF JA] used for Japanese (Kana+Romaji)
[DIFF ZH] used for Mandarin (Hanzi+Pinyin)
[DIFF KO] used for Korean (Hangul+Romaji)
[DIFF ES] used for Spanish
[DIFF IT] used for Italian
[DIFF FR MILLE] used for French
[DIFF TH]* used for Thai (*DL here)

Extra phonemes: [exh], [axh] for exhales
[gs] glotal stops, [cl] closures, [vf] vocal fry

Vocal modes (and Dataset Info):

• Standard (55 min): Standard tone for Nishiren.
• Power (9 min): Stronger and aggressive tone
• Soft (13 min): Darker and gentler tone. Use with adjusted variance parameters to get a whispery voice
• Sweet (7 min): Upbeat and nasaly tone
• Emotional (5 min): Passionate and dramatic tone. Deeper sounding voice
• 2P (6 min): Kayama Gard's voice. Nasaly and strongly voice acted

Multispeaker public datasets: Amaboshi Cipher, OfutonP, PJS, TIGER, Opencpop, M4Singer, CSD, Gianloop's datasets, Ryoku, Petit Millefeuille, Printto TH Dataset

SAMPLE

CHANGELOG

• v2.0 - Switch to multi-dict. Added Korean, Italian, and Thai. Added new Emotional vocal mode. Added breathiness and voicing parameters. Improved sound quality using muon+lynxnet2
• v1.2 - New pitch model using lynxnet. Edited English dictionary with more pronunciation corrections. Finetuned PC-NSF-Hifigan vocoder
• v1.1 - Retrained acoustic and tension with VR. Exported vocoder with mel base e to increase inference speed. Switched f0 extractor from parselmouth to rmvpe, and accelerator to unipc. Recording data added
• v1.02 - Fixed dsdict-ja
• v1.01 - dsvariance file patch
• v1.0 - Full release. Trained with reflow, switched from energy + breathiness to tension, new recording data, added XLS (Mandarin, Spanish, French)
• v0.11 - Updated vocoder
• v0.1 - Initial beta release

CONTACT

For any questions or feedback, please contact Gardanana

• Discord: gardanana
  
• Twitter: @gardanana582
  
• Tumblr: @gardanana
  
• Youtube: @Gardanana