MoeGoe Server

CPU 推理。没有用到 GPU。

Based on Fastapi. Only tts and only support some type model.

Server

Server Main Body Include:

server.py  # main
event.py # lib
config.toml # config

🐾 Tip

The dependencies for this project are based on numpy==1.22.0 , which may break system dependencies!

🪐 Install

pip install -r requirements.txt

apt install libsndfile1

Mkdir model and run server.py to start this server.

After that,Fastapi Docs -> url/docs

🪵 Set Model

Server requirements for model placement

model
|---- somemodel.pth
|---- somemodel.pth.json (== config.json)
|---- info.json

• info.json

Model used for init....

{
  "model": [
    "somemodel.pth"
  ]
}

Param

POST

from pydantic import BaseModel


class TTS_REQ(BaseModel):
    model_name: str = ""
    task_id: int = 1
    text: str = "[ZH]你好[ZH]"
    speaker_id: int = 0
    audio_type: str = "ogg"  # flac wav ogg

RETURN

from pydantic import BaseModel


class TTS_REQ_DATA(BaseModel):
    code: int = 404
    msg: str = "unknown error"
    audio: str = ""
    speaker: str = ""
    model_type: str = ""

OGG

make sure the ogg is encoded with opus codec

Other

• Other Api implementations https://github.com/fumiama/MoeGoe Just found out after writing, SAD

Links_

• MoeGoe_GUI
• Pretrained models

---

How to use

Run MoeGoe.exe

Path of a VITS model: path\to\model.pth
Path of a config file: path\to\config.json
INFO:root:Loaded checkpoint 'path\to\model.pth' (iteration XXX)

Text to speech

TTS or VC? (t/v):t
Text to read: こんにちは。
ID      Speaker
0       XXXX
1       XXXX
2       XXXX
Speaker ID: 0
Path to save: path\to\demo.wav
Successfully saved!

Voice conversion

TTS or VC? (t/v):v
Path of an audio file to convert:
path\to\origin.wav
ID      Speaker
0       XXXX
1       XXXX
2       XXXX
Original speaker ID: 0
Target speaker ID: 6
Path to save: path\to\demo.wav
Successfully saved!

HuBERT-VITS

Path of a hubert-soft model: path\to\hubert-soft.pt
Path of an audio file to convert:
path\to\origin.wav
ID      Speaker
0       XXXX
1       XXXX
2       XXXX
Target speaker ID: 6
Path to save: path\to\demo.wav
Successfully saved!

W2V2-VITS

Path of a w2v2 dimensional emotion model: path\to\model.onnx
TTS or VC? (t/v):t
Text to read: こんにちは。
ID      Speaker
0       XXXX
1       XXXX
2       XXXX
Speaker ID: 0
Path of an emotion reference: path\to\reference.wav
Path to save: path\to\demo.wav
Successfully saved!