ISCA Archive Interspeech 2022
ISCA Archive Interspeech 2022

Conformer Space Neural Architecture Search for Multi-Task Audio Separation

Shun Lu, Yang Wang, Peng Yao, Chenxing Li, Jianchao Tan, Feng Deng, Xiaorui Wang, Chengru Song

Multi-task audio source separation aims to separate the audios collected from the complex environment into three fixed types of signal sources. Existing methods like EAD-Conformer usually take a manually designed model to process the separation. These networks may be sub-optimal since it is hard for humans to train and test all possible architectures. Especially, it is natural to adopt different optimal sub-structures for decoding different types of signals, which, however, is very hard for humans to enumerate. In this paper, we quantitatively analyze the redundancy of the EAD-Conformer network and customize an effective and efficient search space. We propose an efficient K-path search method to search for the optimal architectures from the Conformer-based search space. We conduct a comprehensive search in terms of block numbers, head numbers, and channel numbers. Extensive experiments demonstrate that our searched architectures outperform existing methods in terms of efficiency and effectiveness.


doi: 10.21437/Interspeech.2022-755

Cite as: Lu, S., Wang, Y., Yao, P., Li, C., Tan, J., Deng, F., Wang, X., Song, C. (2022) Conformer Space Neural Architecture Search for Multi-Task Audio Separation. Proc. Interspeech 2022, 5358-5362, doi: 10.21437/Interspeech.2022-755

@inproceedings{lu22b_interspeech,
  author={Shun Lu and Yang Wang and Peng Yao and Chenxing Li and Jianchao Tan and Feng Deng and Xiaorui Wang and Chengru Song},
  title={{Conformer Space Neural Architecture Search for Multi-Task Audio Separation}},
  year=2022,
  booktitle={Proc. Interspeech 2022},
  pages={5358--5362},
  doi={10.21437/Interspeech.2022-755}
}