Our new article by Andronic et al. at the SPECOM conference demonstrates that mp3 coding can help speech recognition systems to become more robust against targeted attacks by adversial noise, see
https://link.springer.com/chapter/10.1007/978-3-030-60276-5_3
