---
tags:
- DiffSVC
- pre-trained_model
- basemodel
- diff-svc
license: "gpl"
datasets:
- 512rc_50k
- 512rc_80k
- 512rc_100k
---
**English** | [简体中文](./README_CN.md)
# DiffSVCBaseModel

A Diff-SVC base model for all kind of voice

## How to use?

1. Choose and download this model 

2. Fill your config and put your datasets into ```(diffsvc-root)/data/raw/{speaker_name}/```

3. Throw this base model(only .ckpt file) into ```(diffsvc-root)/checkpoints/{speaker_name}```

4. Then start preprocessing and training as usual

## How much data do you use?

I use 2 public datasets(opencpop ,m4singer),40h+ audio in total.

## I want to train my own base model!

OK, you can download [this bianry file](./BaseModelBinary.tar.gz).

## Download


| Version        | URL                                  | Reference value of lr |
| -------------- | ------------------------------------ | --------------------- |
| 384rc,50k_step | [Click here](./384rc_50k_step.zip)   | 0.0016                |
| 384rc,80k_step | [Click here](./384rc_80k_step.zip)   | 0.0032                |
| 384rc,100k_step | [Click here](./384rc_100k_step.zip) | 0.0032                |

More coming soon...

## Repos

| Repo            | URL                                                  |
| --------------- | ---------------------------------------------------- |
| Diff-SVC        | [Click here](https://github.com/prophesier/diff-svc) |
| 44.1KHz Vocoder | [Click here](https://openvpi.github.io/vocoders)     |
| M4Singer        | [Click here](https://github.com/M4Singer/M4Singer)   |
| OpenCPOP        | [Click here](https://github.com/wenet-e2e/opencpop)  |

> rc: residual_channels