Discover the Alpha Lexa (Ryan's World) RVC v2 model, retrained for improved performance after 315 epochs with the RMVPE / Mangio-Crepe configuration. This advanced AI voice model captures speech patterns from Ryan's World VTubers channel, offering high-quality text-to-speech output. Generate AI covers or transform text into natural-sounding audio using this unique tool, perfect for creators seeking immersive voice experiences.

Another remake of one of my older models, seemingly been doing a lot of these lately. Dataset is around 4ish minutes of speaking sampled from 5 seperate gaming videos on their "VTubers" channel (which I had to fully watch by the way  ) with BGM and sound effects removed with UVR.

Alpha Lexa (Ryan's World), Retrained

Alpha Lexa (Ryan's World) - RVC v2, 315 Epochs, Retrained, RMVPE / Mangio-Crepe

xhepyxopila

.supers.

currently working on a mangio-crepe version of the model

https://huggingface.co/Xhepyxopila/RyansWorldRVC/resolve/main/AlphaLexaV3.zip

Alpha Lexa (Ryan's World) - RVC v2, 315 Epochs, Retrained, RMVPE, Mangio-Crepe

The singer from those hilariously bad McDonalds Happy Meal ads from Pakistan. Trained off singing data from the Crash Bandicoot, Sonic, and Play Doh ads minus parts with backing vocals. 500 epochs, RMVPE, TITAN pretrain.

Pakistan McDonalds Ad Singer

Trained off voice lines from Wonder and NSMB U since they were the clearest quality. 300 Epochs, OV2 pretrain.

Kamek (Super Mario Franchise, Wii U / 3DS / Switch Era)

Trained off the uncompressed voice files for Star Fox 64 originally found in the 2020 Nintendo gigaleak. Model uses TITAN pretrain with 300 Epochs. Also I made a mistake and forgot to generate an index before training so uhh, I have no clue if doing it after training makes a difference.

Falco Lombardi (Star Fox)

Haven't tested yet but this should theoretically be my best Birdo model to date. Dataset used is Mario Kart 8 Deluxe BCP since it has her lines in the highest quality DeEcho'd and DeVerb'd so the results aren't weirdly robotic sounding RIN-E3 pretrain and 300 epochs.

Birdo (Super Mario Franchise) - The Final Model

Trained off 2 minutes of data with RIN-3 pretrain

Brian Griffin (Family Guy)

Same story as the other retrained RW model, has around half the sample dataset cause my arm was starting to hurt and almost broke during training but SOMEHOW managed to survive and still function.

Alpha Lexa (Ryan's World), Retrained

Description

Add a comment

Combo Panda (Ryan's World), Retrained

Saiba Momoi (Blue Archive)

Jungkook (BTS)

Sonic The Hedgehog [Roger Craig Smith]

ENHYPEN Heeseung

CaseOh

Villager (Minecraft)

FIXED HATSUNE MIKU V3 - VOCALOID [SCROLL DOWN FOR LINK]

Takanashi Hoshino (from Blue Archive)

Satoru Gojo (JJK) [VA Yuichi Nakamura]

Fang [Brawl Stars]

Mortis [Brawl stars]

Melodie - (Brawl Stars) - (English & Korean Voices)

Saiba Momoi (Blue Archive) (VA: Tokui Sora)

Sas (Counter-strike global offensive)

Edgar [Brawl Stars]

SpongeBob SquarePants (Talking And Singing)

Ariana Grande AI

JP Furina / Focalors CV: Inori Minase (Genshin Impact)

Surge (brawl Stars)

JENNIE of BLACKPINK [Strong Ver.]