README.md

---
license: mit
language:
- en
tags:
- RVC
- voice-conversion
- text-to-speech
- voice-cloning
- audio-generation
datasets:
- LJSpeech
- VCTK
metrics:
- MOS (Mean Opinion Score)
- PESQ
- STOI
base_model:
- MangioRVC/Mangio-RVC-Huggingface
pipeline_tag: audio-to-audio
---

# 🌙 LUNAR - High-Quality Female Voice RVC Model

LUNAR is a state-of-the-art **RVC (Retrieval-Based Voice Conversion) model** optimized for **female voice conversion** with studio-grade audio quality at **48kHz sampling rate**. This model delivers natural-sounding voice transformations with minimal artifacts.

##  **Performance & Efficiency Metrics**
Here are the visual benchmarks of Lunar-RVC:

### **1. Training Loss Curve**
![Training Loss](./Metrics/1.jpg)

### **2. Validation Loss Curve**
![Validation Loss](./Metrics/2.jpg)

### **3. Training vs Validation Loss**
![Train vs Val Loss](./Metrics/3.jpg)

### **4. Inference Speed Comparison**
![Inference Speed](./Metrics/4.jpg)

### **5. Audio Quality Scores (MOS)**
![MOS Score](./Metrics/5.jpg)

### **6. GPU Memory Usage**
![GPU Memory](./Metrics/6.jpg)

### **7. Dataset Duration Distribution**
![Dataset Distribution](./Metrics/7.jpg)

### **8. Spectral Convergence**
![Spectral Convergence](./Metrics/8.jpg)

### **9. Model Size Comparison**
![Model Size](./Metrics/9.jpg)

### **10. Efficiency Radar Chart**
![Efficiency Radar](./Metrics/10.jpg)

---

##  Key Features

-  **High-Fidelity Conversion** - Produces natural, expressive female voices
-  **Real-Time Ready** - Optimized for low-latency inference (<20ms/frame)
-  **Pitch & Timbre Control** - Flexible voice modulation capabilities
-  **48kHz Studio Quality** - Professional-grade audio output
-  **Easy Integration** - Compatible with popular voice toolkits

## 📊 Model Specifications

| Parameter          | Value               |
|--------------------|---------------------|
| Framework          | RVC v2             |
| Sample Rate        | 48kHz              |
| Bit Depth          | 16-bit             |
| Model Size         | 1.8GB              |
| Training Hours     | 150 epochs (~10h)  |
| VRAM Requirements  | 4GB+ (inference)   |
| Supported Formats  | WAV, MP3, FLAC     |
---

##  **Inference Guide**
To use Lunar-RVC for inference:

    ``` bash
      # Clone repository
      git clone https://huggingface.co/IssacMosesD/Lunar-RVC-Model
      cd Lunar-RVC
      
      # Install dependencies
      pip install -r requirements.txt
      
      # Run inference
      python infer.py --input input.wav --output output.wav --model Lunar-RVC.pth

##  Use Cases

-  Voice Cloning – Convert your voice into a professional singing voice.

-  Streaming – Real-time voice conversion for content creators.

-  Dubbing – High-quality voice conversion for movies & animations.

-  Music Production – Transform any vocal track into a new singer’s voice.

##  System Requirements

- OS: Windows / Linux

- Python: 3.8+

- GPU: NVIDIA (6GB VRAM minimum recommended)

- CUDA: 11.7+

- Torch: 1.13.1+

##  Contact

- For support, queries, or collaboration:

- Name: [Issac Moses D](https://www.linkedin.com/in/issacmosesd/)

- Email: [issacmsoes19082005@gmail.com](mailto:issacmoses19082005@gmail.com)

- Hugging Face: [IssacMosesD](https://huggingface.co/IssacMosesD)

- GitHub: [Issac Moses D](https://github.com/Issac-Moses)

##  Collaburator

- Name: [Dharani Karnan](https://www.linkedin.com/in/dharani-karnan-060040320/)

- Email: [dharanikarnan18@gmail.com](mailto:dharanikarnan18@gmail.com)

- Hugging Face: [Dharani K](https://huggingface.co/dharzz188)

- GitHub: [Dharani K](https://github.com/Issac-Moses)