File size: 3,032 Bytes
1655e75 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 | ---
license: mit
tags:
- audio
- voice-conversion
- rvc
- safetensors
- maestraea
pipeline_tag: audio-to-audio
---
# RVC Inference Models (Safetensors)
**Retrieval-Based Voice Conversion — V2 Pretrained Models**
[Original Source](https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI) · MIT License
> V2 pretrained models converted from `.pth` to safetensors format (except HuBERT which requires fairseq for deserialization). For use with [Mæstræa AI Workstation](https://github.com/AEmotionStudio/Maestraea).
## What's in This Repo
### Core Models
| File | Size | Description |
|------|------|-------------|
| `hubert_base.pt` | 190 MB | HuBERT feature extractor (kept as .pt — requires fairseq) |
| `rmvpe.safetensors` | 181 MB | RMVPE pitch detection model |
### Pretrained V2 — Generator Models (Inference)
| File | Size | Sample Rate |
|------|------|-------------|
| `pretrained_v2/G32k.safetensors` | 74 MB | 32kHz |
| `pretrained_v2/G40k.safetensors` | 73 MB | 40kHz |
| `pretrained_v2/G48k.safetensors` | 75 MB | 48kHz |
| `pretrained_v2/f0G32k.safetensors` | 74 MB | 32kHz (with F0) |
| `pretrained_v2/f0G40k.safetensors` | 73 MB | 40kHz (with F0) |
| `pretrained_v2/f0G48k.safetensors` | 75 MB | 48kHz (with F0) |
### Pretrained V2 — Discriminator Models (Training)
| File | Size | Sample Rate |
|------|------|-------------|
| `pretrained_v2/D32k.safetensors` | 143 MB | 32kHz |
| `pretrained_v2/D40k.safetensors` | 143 MB | 40kHz |
| `pretrained_v2/D48k.safetensors` | 143 MB | 48kHz |
| `pretrained_v2/f0D32k.safetensors` | 143 MB | 32kHz (with F0) |
| `pretrained_v2/f0D40k.safetensors` | 143 MB | 40kHz (with F0) |
| `pretrained_v2/f0D48k.safetensors` | 143 MB | 48kHz (with F0) |
**Total: ~1.7 GB** (inference-only subset of the full 80 GB RVC repo)
## What RVC Does
RVC (Retrieval-based Voice Conversion) converts vocals from one voice to another:
- **Batch mode** — Upload audio → convert → download result
- **Real-time mode** — Low-latency WebSocket streaming (future)
- Voice models are small (~50–100 MB each) and user-provided
### Key Parameters
| Parameter | Range | Default | Description |
|-----------|-------|---------|-------------|
| `pitch_shift` | -12 to 12 | 0 | Semitone shift |
| `f0_method` | rmvpe/crepe/harvest | rmvpe | Pitch detection |
| `index_rate` | 0–1 | 0.75 | Retrieval index strength |
| `protect` | 0–0.5 | 0.33 | Protect voiceless consonants |
### VRAM Requirements
- **Minimum**: ~2 GB
- **Recommended**: ~6 GB
## Usage with Mæstræa
Place in `~/.maestraea/models/rvc/`. Voice model files (`.pth` + `.index`) go in `~/.maestraea/models/rvc/voices/`.
## License
MIT — same as the original RVC release.
## Credits
- **Model**: [RVC-Project](https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI)
- **Original weights**: [lj1995/VoiceConversionWebUI](https://huggingface.co/lj1995/VoiceConversionWebUI)
- **Conversion & Mirror by**: [AEmotionStudio](https://huggingface.co/AEmotionStudio)
|