Development

Development#

Project Structure#

audio2face/
├── apis/                    # Streaming API implementations
├── data_structures/         # Core data models and Protocol Buffers
├── infer/                   # Inference engines (ONNX and PyTorch)
├── postprocess/             # Postprocessing modules and pipelines
├── service/                 # FastAPI server and error handling
├── split/                   # Audio splitting and segmentation
├── utils/                   # Utility functions and base classes
└── tests/                   # Comprehensive test suite

Testing#

The project includes comprehensive tests with pytest and async testing support. Before running tests, you need to prepare the test input data.

Test Data Preparation#

Download the required test files and organize them in the correct directory structure:

Create test input directory:
```
mkdir -p input
```
Download test files:
- Test Audio: test_audio.wav - Download and save to input/test_audio.wav
- Test Feature: test_feature.npy - Download and save to input/test_feature.npy
Ensure weights directory exists:
```
mkdir -p weights
```
Make sure you have the ONNX model file weights/unitalker_v0.4.0_base.onnx (see Data Preparation section).

Final directory structure should be:

├─audio2face
├─configs
├─docs
├─input
│  ├─test_audio.wav
│  └─test_feature.npy
└─weights
   └─unitalker_v0.4.0_base.onnx

Running Tests#

# Run all tests
pytest tests --log-cli-level=ERROR

# Run specific test categories
pytest tests/apis/           # API functionality tests
pytest tests/infer/          # Inference engine tests
pytest tests/postprocess/    # Postprocessing module tests
pytest tests/data_structures/ # Data structure tests

Test Coverage#

API Tests: Streaming API functionality
Inference Tests: ONNX Unitalker and PyTorch feature extractor validation
Postprocessing Tests: All postprocessing modules with various configurations
Data Structure Tests: FaceClip class functionality and format conversions

Code Quality#

The project maintains high code quality with:

Linting: Ruff for code style and quality checks
Type Hints: Full type annotation support
CI/CD: Automated testing and deployment pipelines