DSANet

Looking and Hearing into Details: Dual-enhanced Siamese Adversarial Network for Audio-Visual Matching

requirments

python 3.8
librosa 0.7.2
numpy 1.19.0
torch 1.4.0
torchvision 0.5.0

Get Dataset and Paper

• Download the VoxCeleb, VGGFace
• Latest Paper List Audio-visual matching

VoxCeleb1

• wav audio data, 1,251 people in total, 39 GB after decompression.

• Baidu Cloud link: VoxCeleb1

• VoxCeleb Document Classification：VoxCeleb1 Document

• Decompression command:

• zip -s 0 split.zip --out unsplit.zip

• unzip unslit.zip

• Vox1 official website: VoxCeleb1

VoxCeleb2

• MP4 video data, files include audio, total of 5,994 people, 255 GB after decompression.

• Baidu Cloud link: VoxCeleb2

• Decompression command:

• zip -s 0 vox2_mp4_dev.zip --out unsplit.zip

• unzip unslit.zip

• Vox2 official website: VoxCeleb2

Contact

If you have any questions, please feel free to contact with me at Netizenwjx@foxmail.com.

Citation

@article{wang2022looking,
  title={Looking and Hearing into Details: Dual-enhanced Siamese Adversarial Network for Audio-Visual Matching},
  author={Wang, Jiaxiang and Li, Chenglong and Zheng, Aihua and Tang, Jin and Luo, Bin},
  journal={IEEE Transactions on Multimedia},
  volume={25},
  pages = {7505-7516},
  year={2023}
}