Skip to content

Latest commit

 

History

History
98 lines (81 loc) · 3.94 KB

make-voicer-VITS2.md

File metadata and controls

98 lines (81 loc) · 3.94 KB

MiriVoice - Make Voicer (VITS2)

📜🧐 : English | 한국어

[EN]

📚 1. Abstract

📚 2. Things to Prepare

❇️ 1. Recorded Samples - recorded.zip

  • First, you have to record samples, by reading Corpus line by line.

🤔What is Corpus?

  • Corpus means a collection of texts. (And its plural form is 'Corpora'.)
  • The Fields that Corpora are used is enormous, but in our case, We will treat the word 'Corpus' as Dataset for TTS Model's train.

🤔What Corpus should I use? How Can I record them?

  • You can use Recstar, or other Recording Software.
  • You can just download one from MiriVoiceSupport-CorpusManager. After finishing your recording, please follow instructions below!
  • If you want, you can make your own dataset.zip yourself.

❇️ After Recording, We will have bunch of wav files.
Please collect each wav files into One Folder, like below:

📂 recorded
├─ 💿MV-KOR-AA-NORMAL_0-001.wav
├─ 💿MV-KOR-AA-NORMAL_0-002.wav
├─ 💿MV-KOR-AA-NORMAL_0-012.wav
├─ 💿MV-KOR-AA-BRIGHT_1-001.wav
├─ 💿MV-KOR-AA-BRIGHT_1-022.wav
└─ ...

Then compress into .zip.

🗂️ recorded.zip
├─ 💿MV-KOR-AA-NORMAL_0-001.wav
├─ 💿MV-KOR-AA-NORMAL_0-002.wav
├─ 💿MV-KOR-AA-NORMAL_0-012.wav
├─ 💿MV-KOR-AA-BRIGHT_1-001.wav
├─ 💿MV-KOR-AA-BRIGHT_1-022.wav
└─ ...

❇️ We're having nice recorded.zip now, all we have to do is making dataset.zip with it.

  • 🔽 Please click below, this notebook will do every dataset.zip stuffs for you!

❇️ 2. Dataset - dataset.zip

  • You will having dataset.zip with below structure, it might be already generated in proper google drive's path.
🗂️ dataset.zip
├─ 📂train
│  ├─ 📜filelist_train.txt.cleaned
│  ├─ 💿MV-KOR-AA-NORMAL_0-001.wav
│  ├─ 💿MV-KOR-AA-NORMAL_0-002.wav
│  ├─ 💿MV-KOR-AA-BRIGHT_1-001.wav
│  └─ ... 
└─ 📂validation
   ├─ 📜filelist_val.txt.cleaned
   ├─ 💿MV-KOR-AA-NORMAL_0-012.wav
   ├─ 💿MV-KOR-AA-BRIGHT_1-022.wav
   └─ ...

❇️ 3. VITS2 Training Notebook

  • 🔽 Click below and run the Colab Notebook.

❇️ 4. VITS2 Voicer Export Notebook

  • After training, we will having G_*.pth, D_*.pth. Please choose one best checkpoint.

  • We cannot use .pth file in MiriVoice. We need to convert it into .onnx format.

  • 🔽 Click below to export your model into MiriVoice's Voicer format.

  • You will now have <voicer name>.zip now, like :

    🗂️ <voicer name>.zip
├─ ⬜voicer.onnx
├─ 📜voicer.yaml
├─ 📜config.yaml
└─ 📜readme.txt

  • You can edit voicer.yaml to add icon or protrait, please see here.

❇️ 5. Install Voicer in MiriVoice

  1. Click Tools -> Install Voicer.
  2. Open your <voicer name>.zip and wait. Your <voicer name>.zip will be extracted in Voicer Folder.
  3. Done! You can use your voicer now.