Audio Feature Extractor

A simple and extensible tool for extracting audio features, designed for speech and audio experiments.

Features

• Extracts traditional audio features (pitch, timbre, loudness, etc.)
• Supports English and Chinese G2P (grapheme-to-phoneme) conversion
• Embedding and semantic feature extraction (e.g., speaker embeddings, ASR, etc.)
• Modular design for easy extension

Installation

1. Clone the repository:

   git clone <your-repo-url>
   cd audio_feature_extractor
   

2. Install dependencies:

   pip install .
   

   Note: For GPU support, install the appropriate version of PyTorch for your CUDA version before running pip install .. See PyTorch Get Started for details.

Usage

from audio_feature_extractor import FeatureExtractor

extractor = FeatureExtractor()
audio_path = "tmp/test.wav"
features = extractor.extract_features(audio_path)
print(features)

Project Structure

src/audio_feature_extractor/
    extractor.py
    features/
        traditional.py
        embedding.py
        semantic.py
    g2p/

Notes

• This tool is intended for research and experimental use.
• For large models or neural network weights, see the documentation for download instructions.
• If you encounter CUDA/cuDNN or dependency issues, please refer to the Troubleshooting section below.

Troubleshooting

• CUDA/cuDNN errors:
  Make sure your environment variables (e.g., LD_LIBRARY_PATH) are set correctly and that you have installed the correct CUDA/cuDNN versions.
• PyTorch version:
  Install the correct PyTorch version for your hardware and CUDA version before installing this package.

License

MIT

Changelog

v0.1.0

• Initial release: basic usage for English speech feature extraction.

Feel free to contribute or open issues!