The key contributions of this work lies in the 'encoding_decoding.py' module, designed to be plug-and-play. Within this module, 'unbiased_encoding' and 'weighted_adjacent_decoding' functions are respectively employed to eliminate quantization errors during the training and decoding phases in deep learning-based sound source localization.
If you found this repo is helpful, please consider citing our paper 😊
@article{feng2024eliminating,
title = {Eliminating quantization errors in classification-based sound source localization},
author = {Linfeng Feng and Xiao-Lei Zhang and Xuelong Li},
journal = {Neural Networks},
volume = {181},
pages = {106679},
year = {2025},
issn = {0893-6080},
doi = {https://doi.org/10.1016/j.neunet.2024.106679},
}