---
license: openrail
language:
- en
pipeline_tag: text-to-speech
---

<img src="https://huggingface.co/OtisAI/NORA/resolve/main/NORA.jpg" />

Thank you for viewing my first model! This is a TTS (GPT-soVITS) and RVC train of one of my favorite video game characters, NORA, from Atomic Heart.

# Important Note
- None of the audio, content, or software used to train this model is my property. I do not own the rights to the original voice lines used to train NORA-TTS. All vocal file rights belong to Mundfish and their respective publishers and distributers.

## Model Details
- **Developed by:** otisai
- **Model Type:** Audio Transformer | RVC
- **Model Description:** NORA, inspired by NORA from Atomic Heart, is my first iteration of my incoming series of RVC and TTS models. 
- **License:** OpenRAIL

### Training Config - TTS

NORA was trained in ~14 minutes on a single NVIDIA H200 with 20 epochs on each weight (40 total) and a batch size of 4.

| Configuration Item    | NORA-TTS                |
|-----------------------|-------------------------|
| **GPU**               | NVIDIA H200             |
| **Dataset**           | 25 Minutes of Audio     |
| **Total Epochs**      | 40                      |
| **SoVITS Epochs**     | 20                      |
| **GPT Epochs**        | 20                      |
| **Learning Rate**     | 0.4                     |
| **Batch Size**        | 4                       |

### Training Config - RVC

NORA was trained in a 1.5-hour period on a single NVIDIA H200 at 1000 total epochs with a batch size of 8. 

| Configuration Item    | NORA                    |
|-----------------------|-------------------------|
| **GPU**               | NVIDIA H200             |
| **Dataset**           | 25 Minutes of Audio     |
| **Total Epochs**      | 1000                    |
| **Batch Size**        | 8                       |

## Collaborators

- [otisai](https://huggingface.co/OtisAI)