RVC Python

A Python implementation for using RVC (Retrieval-based Voice Conversion) via console, Python scripts, or API.

Table of Contents

• Features
• Installation
• Usage
  • Command Line Interface
  • Python Module
  • API
• Model Management
• Options
• Advanced Setup (GPU Acceleration)
• Changelog
• Contributing

Demo

bandicam.2023-12-27.01-36-03-212.mp4

Features

• Console interface for single file or batch processing
• Python module for integration into other projects
• API server for remote processing
• Support for both CPU and GPU acceleration
• Dynamic model loading and unloading
• Flexible model directory management

Installation

Basic Installation (CPU only)

pip install rvc-python

Recommended Installation (with GPU support)

For Windows:

python -m venv venv
venv\Scripts\activate
pip install rvc-python
pip install torch==2.1.1+cu118 torchaudio==2.1.1+cu118 --index-url https://download.pytorch.org/whl/cu118

For Linux:

python -m venv venv
source venv/bin/activate
pip install rvc-python
pip install torch==2.1.1+cu118 torchaudio==2.1.1+cu118 --index-url https://download.pytorch.org/whl/cu118

Usage

Command Line Interface

The CLI supports two modes: cli for direct file processing and api for starting an API server.

python -m rvc_python [-h] {cli,api} ...

CLI Mode

Process a single file or batch of files:

python -m rvc_python cli -i INPUT -o OUTPUT -mp MODEL [options]

Example:

python -m rvc_python cli -i input.wav -o output.wav -mp path/to/model.pth -de cuda:0

API Mode

Start the API server:

python -m rvc_python api [-p PORT] [-l] [options]

Example:

python -m rvc_python api -p 5050 -l

Python Module

from rvc_python.infer import RVCInference

rvc = RVCInference(device="cuda:0")
rvc.load_model("path/to/model.pth")
rvc.infer_file("input.wav", "output.wav")

API

The API server provides several endpoints for voice conversion and model management. Here's a detailed breakdown of each endpoint:

1. Convert Audio

• Endpoint: POST /convert
• Description: Converts an audio file using the currently loaded model.
• Request Body:

  {
    "audio_data": "base64_encoded_audio"
  }

• Response: The converted audio file (WAV format)
• Example:

  import requests
  import base64
  
  url = "http://localhost:5050/convert"
  with open("input.wav", "rb") as audio_file:
      audio_data = base64.b64encode(audio_file.read()).decode()
  
  response = requests.post(url, json={"audio_data": audio_data})
  
  with open("output.wav", "wb") as output_file:
      output_file.write(response.content)

2. List Available Models

• Endpoint: GET /models
• Description: Returns a list of all available models.
• Response: JSON array of model names
• Example:

  response = requests.get("http://localhost:5050/models")
  models = response.json()
  print("Available models:", models)

3. Load a Model

• Endpoint: POST /models/{model_name}
• Description: Loads a specific model for use in conversions.
• Response: Confirmation message
• Example:

  response = requests.post("http://localhost:5050/models/my_model")
  print(response.json())

4. Get Current Parameters

• Endpoint: GET /params
• Description: Retrieves the current parameter settings.
• Response: JSON object with current parameters
• Example:

  response = requests.get("http://localhost:5050/params")
  print("Current parameters:", response.json())

5. Set Parameters

• Endpoint: POST /params
• Description: Updates the parameters for voice conversion.
• Request Body:

  {
    "params": {
      "f0method": "harvest",
      "f0up_key": 0,
      "index_rate": 0.5,
      "filter_radius": 3,
      "resample_sr": 0,
      "rms_mix_rate": 0.25,
      "protect": 0.33
    }
  }

• Response: Confirmation message
• Example:

  params = {
    "f0method": "harvest",
    "f0up_key": 2,
    "protect": 0.5
  }
  response = requests.post("http://localhost:5050/params", json={"params": params})
  print(response.json())

6. Upload a New Model

• Endpoint: POST /upload_model
• Description: Uploads a new model (as a zip file) to the server.
• Request: Multipart form data with a zip file
• Response: Confirmation message
• Example:

  with open("new_model.zip", "rb") as zip_file:
      files = {"file": ("new_model.zip", zip_file)}
      response = requests.post("http://localhost:5050/upload_model", files=files)
  print(response.json())

7. Set Computation Device

• Endpoint: POST /set_device
• Description: Sets the device (CPU/GPU) for computations.
• Request Body:

  {
    "device": "cuda:0"
  }

• Response: Confirmation message
• Example:

  response = requests.post("http://localhost:5050/set_device", json={"device": "cuda:0"})
  print(response.json())

Model Management

Models are stored in the rvc_models directory by default. Each model should be in its own subdirectory and contain:

• A .pth file (required): The main model file.
• An .index file (optional): For improved voice conversion quality.

Example structure:

rvc_models/
├── model1/
│   ├── model1.pth
│   └── model1.index
└── model2/
    └── model2.pth

You can add new models by:

1. Manually placing them in the rvc_models directory.
2. Using the /upload_model API endpoint to upload a zip file containing the model files.
3. Using the /set_models_dir API endpoint to change the models directory dynamically.

Options

Input/Output Options

• -i, --input: Input audio file (CLI mode)
• -d, --dir: Input directory for batch processing (CLI mode)
• -o, --output: Output file or directory

Model Options

• -mp, --model: Path to the RVC model file (required for CLI, optional for API)
• -md, --models_dir: Directory containing RVC models (default: rvc_models in the current directory)
• -ip, --index: Path to the index file (optional)
• -v, --version: Model version (v1 or v2)

Processing Options

• -de, --device: Computation device (e.g., "cpu", "cuda:0")
• -me, --method: Pitch extraction method (harvest, crepe, rmvpe, pm)
• -pi, --pitch: Pitch adjustment in semitones
• -ir, --index_rate: Feature search ratio
• -fr, --filter_radius: Median filtering radius for pitch
• -rsr, --resample_sr: Output resampling rate
• -rmr, --rms_mix_rate: Volume envelope mix rate
• -pr, --protect: Protection for voiceless consonants

API Server Options

• -p, --port: API server port (default: 5050)
• -l, --listen: Allow external connections to API server
• -pm, --preload-model: Preload a model when starting the API server (optional)

Changelog

For a detailed list of changes and updates, please see the Releases page.

Contributing

Contributions are welcome! Feel free to submit pull requests or open issues for bugs and feature requests.