Data Preparation

Data Preparation#

To use Audio2Face, you need to download the ONNX model file and set up the required directory structure.

Download ONNX Model#

Step 1: Download the Model File#

Download the ONNX model file from one of the following sources:

Step 2: Organize the Data#

  1. Create the weights directory:

    • Create a weights directory in your project root if it doesn’t exist

    • This directory will store the ONNX model files

  2. Place the model file:

    • Place the downloaded unitalker_v0.4.0_base.onnx file in the weights directory

    • Ensure the file is named exactly unitalker_v0.4.0_base.onnx

  3. Verify directory structure:

    • Ensure the following directory structure is created:

├─audio2face
├─configs
├─docs
└─weights
   └─unitalker_v0.4.0_base.onnx

Directory Structure#

The following directory structure is required for Audio2Face to function properly:

  • weights/: A folder for storing ONNX model files

    • This directory must exist in the project root

    • All model files should be placed in this directory

  • weights/unitalker_v0.4.0_base.onnx: The main ONNX model file for audio-to-face conversion

    • This is the core inference model used by the service

    • The file must be placed in the weights directory

    • Ensure the file is not corrupted and is the correct version (v0.4.0)

Verification#

After downloading and organizing the data, you can verify the setup:

# Check if the weights directory exists
ls -la weights/

# Verify the model file is present
ls -la weights/unitalker_v0.4.0_base.onnx

Next Steps#

Once you have completed the data preparation:

  1. Follow the Quick Start: See Quick Start for detailed quick start instructions

  2. Follow the Installation Guide: See Installation Guide for detailed environment setup instructions

  3. Local Development: Set up your local development environment as described in the installation guide

Contents