Model Maker Role

Last update: February 28, 2026

---

To upload your Voice Models in the AI HUB Discord Server's #voice-models forum channel, you need to pass a Quality Control (QC) check to be sure that you post good voice models.

---

‎

It lacks the correct files.

• The .ZIP file must contain both the correct .INDEX & .PTH file.

• The correct .index is the one named added_.

  • The added index contains the voice's accent and speech manor.

• The correct .pth is the one that has your model's name, for example: TylerSwift_e60_s120.pth

  • The .pth contains the actual model and pitch data.

---

‎

Model is low quality.

• A bad model:
  • Sounds scratchy/screechy.
  • Has a muffled sound.
  • Sounds inaccurate to the source.
  • Is incapable of hitting certain notes.
  • Has slurred speech.
  • Is unable of pronouncing words correctly in its intended language.
  • Has artifacting.
  • Has noise.

---

‎

An outdated extraction method was used.

• Only Crepe & RMVPE are allowed. Learn about them here

• Harvest, Dio, Crepe-Tiny, PM, etc. are obsolete.

---

‎

The audio demo contains instrumental.

• Don't include ANY music in the audio demo, even if it's not copyrighted. This is due to:
  • Concerns over copyright.
  • In many cases, the music can "hide" the flaws of the voice model, making it harder to judge its quality.

---

The audio demo is altered.

• Don't add reverb, equalize, or alter the demo in any way, as it won't be a faithful representation of the model. It must be the raw, unmodified output from the inference.

• Trimming silences at the beginning/end of the audio demo is allowed.

‎

---

Is a robotic or non-human voice.

• Robotic, sound effect and drum models will also be rejected, because with these types of voices it is difficult to determine if you know how to clean a dataset properly.

• However once you get model maker you will be able to post robotic, sound effect or drum models.

‎

---

Step 1: Prepare the submission.

• Once your model is ready, head over to the AI HUB's #model-maker-role channel.

• Click the Submit Model button.

‎
‎

model-name

Its name.

technology

The technology used for its training:

• RVC
• GPT-SoVITS

extraction

The extraction method you used:

• RMVPE
• Crepe
• Mangio-Crepe (Obsolete)
• Harvest (Obsolete)
• PM (Obsolete)
• DIO (Obsolete)

vocoder

The vocoder you used:

• HifiGan
• RefineGan

epochs

Total epochs amount.

---

Step 2: Complete the submission.

• You will get a DM by Wally asking for you to Complete the Submission.

• Click the Complete Submission button.

‎
‎

Embedder

The Embedder Model you used:

• ContentVec
• Spin
• SpinV2

Pretrain

The Pretrain you used.

Model Link

Its download link from Hugging Face (right click and copy link the download icon of the model zip, for example https://huggingface.co/Nick088/TADC_Bubble/resolve/main/TADC_Bubble.zip?download=true) or Weights.

Sample File

An audio sample of it talking/singing.

Additional Information

Optional. Add more context about the model if you want.

• Click the Submit button.

---

---

‎

Step 4: Send submission.

• Once you are done filling the information it will send your model to get QCed

  ‎ ‎

• Now, your model will be posted in #model-maker-submissions where other model makers will upvote or downvote your model reviewing it. After one week, the model maker submission will be accepted or rejected based on voting.

• If you made a mistake in your submission or you want to change something, you can try to contact staff or talk about in the #model-maker-role discussion.

• If your model gets approved, you can then repost the model (& future models) to the #voice-models forum.

---

‎