The Next-Generation of Virtual Divas

Eternally Iconic

Entirely Ethical AI Vocal Synthesis.

NEODIVA // ETHICAL AI SYNTHESIS // ICONIC // COMMUNITY DRIVEN // DIFFSINGER ENGINE // FUTURE SOUND // NEODIVA // ETHICAL AI SYNTHESIS // ICONIC // COMMUNITY DRIVEN // DIFFSINGER ENGINE // FUTURE SOUND //

01

Who Are We?

NeoDIVA is a group dedicated to creating unique AI Vocalists, utilising OpenVPI's DiffSinger engine. We are pushing the boundaries of what virtual idols can be.

02

Lead Vocals

NeoDIVA's "Lead" vocals are designed to provide the essential foundations for AI vocal synthesis, offering a range of voices that includes both masculine and feminine options, as well as a neutral or in-between voice.

03

Collaboration

At NeoDIVA, we believe in the strength of community and uplifting others. We are committed to collaborating, supporting creators, and expanding our lineup. If you'd like to collaborate, please submit your vocal for consideration as a NeoDIVA feature.

Affiliates

Core Team

AlexFloareaVT

/// Project Lead

Lead Developer & Project Lead. Alex is the mastermind behind the NeoDIVA project.

Wormwood_PROD

/// Audio & Art

Key Artist and Music Producer. They assist in creating illustrated assets for NeoDIVA.

Algernyan

/// Design

Character Designer. They assist in creating unique designs for the NeoDIVA lead vocals.

VOCALISTS

CHOOSE YOUR VOCALIST

Lead Series

Alex Floarea

/// UNDEAD IDOL

Irone Aka

/// TIME TRAVELER

WURMLO1D

/// WASTELAND BOT

Feature Series

Kuro Watanabe

Essaurus

XENA

Ami

Biography

Language

Data Size

Rec. Range

Rec. Tempo

Download Voicebank

/// Production:

/// Multi-Language Data: Tigermeat, PJS, Criss, Lottie, PrintMOV, Ariika, CSD, Gianloop, Netriko, Adrianna Stan, PixPrucer, Team BRAPA

Audio Samples

RVC PROTOCOL

/// What is RVC?

"Retrieval-based Voice Conversion (RVC) is an open source voice conversion AI algorithm that enables realistic speech-to-speech transformations, accurately preserving the intonation and audio characteristics of the original speaker" (Wikipedia).

This can be used within the context of UTAU to easily create new voicebanks, without needing to record new data.

/// What do we offer?

NeoDIVA currently offers the following services relating to RVC, free of charge:

RVC Model Training
RVC to UTAU Conversion (Currently available in English Arpasing, JPN VCV and CVVC, and French CVVC.)

/// Interested in working with us?

If you are interested in working with us for our RVC services please make sure that you read the T.O.S before hand, which can be found here.

If you have read through the T.O.S and agree to it you can register your interest with our submissions form, which can be found here.

If you are interested in volunteering your UTAU voicebank to be used as inference data for this service, please contact us directly either via email or through the NeoDIVA twitter account.

Resources

Alex Floarea 2.0 Dataset

Transcriptions & CSV

Download

The "Alex Floarea Public Dataset" is shared with a CC BY-NC 4.0 license. You are free to use this dataset in any non-commercial project as long as the resulting embeddings for "Alex Floarea" are not publicly available and "NeoDIVA" must be credited. Special exceptions to both of these terms may be granted on a case-by-case basis, with a licensing fee potentially being charged. This does not apply to selling original music created with a model that uses this dataset as support data, as long as the voice of Alex Floarea is not the vocalist used.

The use of RVC, or any other SVC engine, is prohibited both for Inference and training an SVC model of this voice. As above special exceptions may be granted, and the resulting data and/or model will be subject to the same license as the original dataset.

Romanian G2P DiffSinger Phonemizer

PLUGIN // OPENUTAU

Download

NeoDIVA .lab Phoneme Error Checker

UTILITY // SCRIPT

Download

F.A.Q.

What exactly are "NeoDIVAS" ?

"NeoDIVA"s are DiffSinger voicebanks trained in our NeoDIVA model.

What is the difference between lead and feature vocals?

NeoDIVA "LEAD" vocals are our own personal collection of DiffSinger voicebanks, whereas "FEATURE" vocals are community-submitted vocalists that we have trained.

How is this ethical?

Our models are trained only on vocal data provided with explicit consent from the vocalists for this specific purpose. We never scrape or use data without permission. The NeoDIVA model is also trained locally, rather than through an unsustainable data centre, and only when updates are needed.

How many languages do they support?

Currently, the NeoDIVA models support 13 languages:

English, Japanese, Thai, Korean, Italian, Spanish, Chinese, Portuguese (Brazilian), Georgian, Romanian, Polish, Russian and French

Will the NeoDIVA team label my data?

The NeoDIVA team will not be able to label your data for you, but we will offer assistance where possible.

Will I keep the rights to my character?

If accepted as a NeoDIVA "FEATURE" vocalist, you will retain 100% of the rights to your character and/or dataset.

Can my Neodiva Diffsinger be used commercially?

Unfortunately, due to some of the licenses of the datasets used in our model, commercial usage is NOT allowed for any NeoDIVA vocals. This may change in the future, and you will be notified of any changes. If this does happen, the option to remain without a commercial license is still available.

vocalist Submission

● Disclaimer

By completing this form, you acknowledge and agree that your dataset will be considered for training as a NeoDIVA Feature. As such, they will be exclusively featured in this capacity and will not serve as a "Lead" vocal, as they represent a distinct group used for promoting NeoDIVA as a community.

It is mandatory that users read the Terms of Service before submitting. Read it HERE

The DiffSingers showcased on the NeoDIVA site belong to and are managed by their individual creators. Our platform acts as a catalyst for DiffSinger voicebanks and character growth. We encourage you to submit your vocalists for careful evaluation and possible inclusion in our handpicked collection.

Launch Form