PLCPA-ASYM loss

The power-law compressed phaseaware (PLCPA) loss

The definition is as follows:

$$ \mathcal{L}_{PLCPA}=\left||S|^p-|\hat{S}|^p\right|^2+\left||S|^pe^{j\varphi(S)}-|\hat{S}|^pe^{j\varphi(\hat{S})}\right|^2 $$

The definition of asymmetric loss is as follows:

Citation

@article{eskimez2021human,
  title={Human listening and live captioning: Multi-task training for speech enhancement},
  author={Eskimez, Sefik Emre and Wang, Xiaofei and Tang, Min and Yang, Hemin and Zhu, Zirun and Chen, Zhuo and Wang, Huaming and Yoshioka, Takuya},
  journal={arXiv preprint arXiv:2106.02896},
  year={2021}
}

@misc{eskimez2021personalized,
  title={Personalized Speech Enhancement: New Models and Comprehensive Evaluation}, 
  author={Sefik Emre Eskimez and Takuya Yoshioka and Huaming Wang and Xiaofei Wang and Zhuo Chen and Xuedong Huang},
  year={2021},
  eprint={2110.09625},
  archivePrefix={arXiv},
  primaryClass={eess.AS}
}