LMAC-TD: Producting Producing Time Domain Explanations for Audio Classifiers

This code is extracted from a private fork of Speechbrain. We will open a PR to the official repo upon acceptance of the paper.

Classifier

---

Information to re-train the classifier are in the classification folder. The classifier checkpoints are automatically downloaded for interpreter training. That is, to reproduce the results of the paper you do not need to re-train the classifier. We provide the code for completeness.

Interpreter

---

All the code relevant to train the interpreter is available in interpret folder. We invite you to check out the folder's README.md for instructions to reproduce the training and evaluation.

Referencing LMAC-TD

If you use our work in your research, please cite

@inproceedings{mancini2025lmactd,
  author={Mancini, Eleonora and Paissan, Francesco and Ravanelli, Mirco and Subakan, Cem},
  booktitle={ICASSP 2025 - 2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)}, 
  title={LMAC-TD: Producing Time Domain Explanations for Audio Classifiers}, 
  year={2025},
  pages={1-5},
  doi={10.1109/ICASSP49660.2025.10890448}}