End-to-End Multi-Loss Training for Low Delay Packet Loss Concealment

Nan Li, Xiguang Zheng, Chen Zhang, Liang Guo, Bing Yu

Real-time teleconferencing has become one of the essential parts in our daily life. While packet loss during real-time data transmission is unavoidable, traditional signal processing based Packet Loss Concealment (PLC) techniques have been developed in recent decades. In recent years, deep learning based approaches have also proposed and achieved state-of-the-art PLC performance. This work presents a low-delay multi-loss based neural PLC system. The multi-loss is consisted by a signal loss, a perceptual loss and an ASR loss ensuring good speech quality and automatic speech recognition compatibility. The proposed system was ranked 1st place in INTERSPEECH 2022's Audio Deep Packet Loss Concealment Challenge.

doi: 10.21437/Interspeech.2022-11439

Cite as: Li, N., Zheng, X., Zhang, C., Guo, L., Yu, B. (2022) End-to-End Multi-Loss Training for Low Delay Packet Loss Concealment. Proc. Interspeech 2022, 585-589, doi: 10.21437/Interspeech.2022-11439

@inproceedings{li22ea_interspeech,
  author={Nan Li and Xiguang Zheng and Chen Zhang and Liang Guo and Bing Yu},
  title={{End-to-End Multi-Loss Training for Low Delay Packet Loss Concealment}},
  year=2022,
  booktitle={Proc. Interspeech 2022},
  pages={585--589},
  doi={10.21437/Interspeech.2022-11439}
}