Install

Dependencies

  1. Download Anaconda

  2. Create a conda environment

# models were trained using python==3.9.6
conda create -n myCondaEnvironment python=3.9.6
  1. Install PyTorch with or without CUDA support. This library was developed with torch==1.12.0.

  2. Download the aiSFX package with either method

# Using the released pip package
pip install aisfx
# Using the latest updated code
pip install git+https://github.com/alisonbma/aiSFX.git
  1. To replicate the ISMIR paper code, you will need to download the Essentia library to compute the spectrograms.


pip install essentia==2.1b6.dev778

Operating System Compatibility

Warning

  • As of the moment, Essentia does support installation on MacOS. However, installation is only supported on Homebrew, which does not work well with conda environments. Thus, to re-create our results, we suggest you use Linux or WSL on Windows. If you still wish to do so, refer to the API Reference and use aisfx.inference.main() as a reference for any modifications to the embedding extraction code.

  • For convenience, we have also implemented an option to run the code with Librosa, see aisfx.spectrogram.spectrogram_librosa.compute_spectrogram(). This may produce lower results.