Wisdom King Raphael (智慧之王 ラファエル) — RVC v2 Voice Model
RVC v2 • 40 kHz • RMVPE • 200 Epochs • 3,400 Steps • Clean ~3 min Dataset • Applio
An RVC v2 (Retrieval-based Voice Conversion) model trained to faithfully recreate the calm, analytical, and majestic voice of Wisdom King Raphael (智慧之王 ラファエル) from That Time I Got Reincarnated as a Slime (Tensei Shitara Slime Datta Ken).
Built from a manually cleaned speech dataset and trained using Applio (RVC v2), emphasizing natural articulation, stable pitch inference, and consistent character identity.
Designed for integration with Raphael OS, conversational AI agents, and interactive voice applications.
Audio Demo
Sample generated using NanamiNeural (TTS Base) converted to Wisdom King Raphael:
Base Voice: Microsoft Edge TTS (ja-JP-NanamiNeural)
Converted Voice: Wisdom King Raphael (RVC v2 Model)
「告。ユニークスキル『叡智之者』は、アルティメットスキル、ラファエルに進化しました。」
(Notice. Unique Skill 'Eichiaru Mono' has evolved into Ultimate Skill, Raphael.)
Training & Model Details
| Parameter | Specification |
|---|---|
| Character Target | Wisdom King Raphael (智慧之王 ラファエル) |
| Dataset Quality | ~3 Minutes of Manually Curated Speech Dataset |
| Dataset Slices | ~150 - 200 Clean Audio Slices (4s segments) |
| Model Architecture | RVC v2 (Retrieval-based Voice Conversion) |
| Training Framework | Applio (RVC v2) |
| Pitch Extraction Algorithm | RMVPE |
| Sample Rate | 40,000 Hz (40k) |
| Total Epochs | 200 Epochs |
| Total Training Steps | 3,400 Steps |
| Pretrained Base Model | HiFi-GAN (f0G40k / f0D40k) |
Compatibility
Tested and compatible across major RVC ecosystems:
- ✅ Applio
- ✅ RVC WebUI v2
- ✅ Mangio-RVC
- ✅ w-okada Realtime Voice Changer
Repository Files
| File Name | Description | Size |
|---|---|---|
Raphael_200e_3400s.pth |
Trained RVC Model Weights | 53.9 MB |
Raphael.index |
Feature Retrieval Index File | ~1.2 MB |
Recommended Inference Settings (Applio / RVC WebUI)
For optimal and natural Raphael voice conversion:
Pitch Extraction Method: rmvpe
Index Rate: 0.8
Protect Voice Consonants: 0.33
Clean Audio: True (Strength: 0.5)
Resample Output: 0
TTS Base Voice (Japanese): ja-JP-NanamiNeural
TTS Base Voice (English): en-US-AvaNeural / en-US-GuyNeural
Pitch Key Offset: 0 (or +12 when converting from low male voice)
Usage Notes
- Optimized primarily for high-fidelity Japanese voice conversion.
- English voice conversion quality depends on the source speaker's pronunciation and articulation.
- Use clean, noise-free input audio for optimal voice similarity and pitch accuracy.
Limitations
- Performance is optimized for clean, noise-free speech input.
- Cross-language voice conversion (e.g., non-Japanese input) may produce varying pronunciation quality.
- Extreme pitch shifts may reduce naturalness and speaker identity similarity.
- This model is intended solely to emulate a fictional character's voice for non-commercial purposes.
Integration via CLI / Python
Download via Wget:
# Download Weights
wget -O assets/weights/Raphael.pth "https://huggingface.co/zidanaetrna/wisdom-king-raphael/resolve/main/Raphael_200e_3400s.pth"
# Download Index
wget -O logs/Raphael/Raphael.index "https://huggingface.co/zidanaetrna/wisdom-king-raphael/resolve/main/Raphael.index"
Download via huggingface_hub Python SDK:
from huggingface_hub import hf_hub_download
pth_path = hf_hub_download(
repo_id="zidanaetrna/wisdom-king-raphael",
filename="Raphael_200e_3400s.pth"
)
index_path = hf_hub_download(
repo_id="zidanaetrna/wisdom-king-raphael",
filename="Raphael.index"
)
print(f"Model downloaded to: {pth_path}")
Notice & License
- License: Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0).
- Usage Scope: This model is intended for non-commercial research, educational use, and personal projects in accordance with the CC BY-NC 4.0 license (Raphael OS Project).
- Credits: Voice dataset manually curated and trained by Zidan Mutaqin. Character rights belong to Fuse and Micro Magazine / TenSura Production Committee.
Developed by Zidan Mutaqin |
GitHub
「告。解析を完了しました。」