asr-visualization

The purpose of this repository is to explore ASR visualization in terms of weighted attention relevance and weighted convolution relevance. The findings of this research has been submitted to ICASSP 2024.

References

@electronic{Weinberger2015,   
    title = {{Speech Accent Archive}},   
    url = {http://accent.gmu.edu},   
    author = {Steven Weinberger},   
    year = {2015},
    keywords = {accent}
}
@article{Sun2023,
  author={Sun, Tianli and Chen, Haonan and Hu, Guosheng and He, Lianghua and Zhao, Cairong},
  journal={IEEE Transactions on Multimedia}, 
  title={Explainability of Speech Recognition Transformers via Gradient-based Attention Visualization}, 
  year={2023},
  volume={14},
  number={8},
  pages={1-12},
  doi={10.1109/TMM.2023.3282488}
}

Shield:

This work is licensed under a Creative Commons Attribution 4.0 International License.