Dunder_CV+ and Lain Noumen+ released for Diffsinger!

Sad Boy and Mask Man take on a new era!

Hello Everyone! Following Today’s release of Lain Noumen Natural for OpenUTAU, we have finished the steps needed to releases Lain Noumen+ ~Proof of Concept~ (referred to as Lain Noumen Natural+ in a prior article)and Dunder_CV+ ~Proof of Concept~. These voicebanks are considered betas due to the fact their English is a bit buggy on end notes as they used auto-labeled English data.

Voicebank

For starters, Lain Noumen+ ~Proof of Concept~ was trained on samples of Lain Noumen Natural with the consent of his voice provider, and Dunder_CV+ ~Proof of Concept~ was trained on samples of Dunder_CV Act II voicebanks (consent obviously given). During training they were accompanied by 4 folders of the voice color data being used in Dunder_Intercalaris_DS (name subject to change), these colors were Intercalaris (Ported from ENUNU, also the only manually labeled data), Mercedonius (Ported from ENUNU), Rap, and Switchy (data that switches from JP and EN in quick succession), which have enabled greater range and English capabilities in the UTAU ports.
When Dunder_Intercalaris_DS’s English data is manually labeled we will retrain these voicebanks to make English less buggy. Currently, English end phonemes tend to slur into other phonemes, and vowels can act weird in isolation. Most of this can be fixed by just adjusting phoneme timing, but we still want to fix it in the future.

Dunder_CV+ ~Proof of Concept~ has five voice colors, one for each voicebank in the original. Below are samples of each, we have not adjusted phoneme timing for the sake of presenting these voicebanks as they are out of the box.

Just One More Time – Aki
Just One More Time – Fuyu
Just One More Time – Genesis
Just One More Time – Haru
Just One More Time – Natsu

Lain Noumen+ ~Proof of Concept~ has 3 voice colors. Default Tone which was trained on pure Lain Noumen Natural outputs, Natural Tone which was trained on outputs of C3 pitch only, and High Tone which was trained on outputs of G3 pitch only. Again for the sake of presenting the voicebanks exactly as they are, phoneme timing has not been adjusted:

Just One More Time – High Tone
Just One More Time – Default Tone
Just One More Time – Natural Tone

There are differences in the sound of all 3 but they’re not super distinct from each other. If Nobody (alias) wants to record more appends for Lain in the future, we will update Lain Noumen+ ~Proof of Concept~ to include these as long as he consents.

Design

The direction we’ve chosen for the plus voicebanks image wise is sort of a riff on SynthV gold series, gold stripe included. It’s just their normal designs with a yellow-orange gradient multiplied over them. Dunder_CV+ ~Proof of Concept~ Icon differs from Dunder_CV Act II Complete, in that it only features the art for Dunder_Haru Act II, as Dunder needing a power English vocal is sort of the reason these plus voicebanks exist to begin with.
I’m also thinking of giving the two unique art (and maybe designs if I feel like it) but that will happen when I don’t have internship responsibilities.

License Updates

Lain Noumen, Dunder, and Ascenia.EXE’s licenses have been updated to a new handwritten one. If you were considering purchasing these licenses please take a gander at the latest versions to see what may have changed (as of date nobody has bought their original licenses so a massive change like this is fine). Lain Noumen and Dunder also received an additional update today to account for the new voicebanks.

What’s Next?

When I have time I will probably be writing original demo songs for Lain Noumen+ and Dunder_CV+ to properly showcase what their English can sound like with phoneme adjustment, tuning and mixing.
I am working on an Ascenia plus voicebank, but I’m wasn’t super happy with how she came out during training and think I’ll have to redo her OTO first. It might get uploaded today if I change my mind, if not though it’ll probably also have an append using data from her Z Idol Talker voicebank, in addition to Normal and Error. Might even get crazy and give ㄬomeder, Ethel, and Ambrose plus banks. (ㄬomeder has an un-oto-d voicebank but Ethel and Ambrose would probably need to get CV voicebanks at least)
I really like weird sounding voicebanks over natural ones so I want to experiment more with the kind of thing we’re doing with the plus voicebanks.

That’s all for now, Bye Bii~