/ISSL

Iterative Sound Source Localization

Primary LanguagePython

ISSL

The official implementation for the paper "Iterative Sound Source Localization for Unknown Number of Sources".

Dataset

See https://github.com/FYJNEVERFOLLOWS/VCTK-SIM

Dependency

  • torch+torchaudio <https://pytorch.org/>

  • apkit <https://github.com/hwp/apkit>_ (version 0.2)

  • scikit-learn

  • scipy

Training

  1. Run train_ssnet.py and save a checkpoint of SSNet;
  2. Load the saved SSNet's path and run inference_sps.py on another training set to build the spatial spectrum dataset;
  3. Run train_asdnet.py on the spatial spectrum dataset and save a checkpoint of ASDNet;
  4. Load both SSNet and ASDNet, run inference.py to test DOA estimation performance on unknown multiple sources.