Distributed Summation Privacy for Speech Enhancement

Matt O’Connor, W. Bastiaan Kleijn


Speech privacy in modern sensor network environments is necessary for widespread adoption and public trust of collaborative acoustic signal processing. Most current distributed privacy research deals with ensuring local node observations are not accessible by neighbouring nodes while still solving shared tasks. In this work we develop the concept of distributed task privacy in unbounded public networks, where linear codes are used to create limits on the number of nodes contributing to a distributed summation task, such as beamforming. We accomplish this by wrapping local observations in a linear code and intentionally applying symbol errors prior to transmission. If many nodes join a distributed speech enhancement task, a proportional number of symbol errors are introduced into the aggregated code leading to decoding failure if the code’s predefined symbol error limit is exceeded.


 DOI: 10.21437/Interspeech.2020-1977

Cite as: O’Connor, M., Kleijn, W.B. (2020) Distributed Summation Privacy for Speech Enhancement. Proc. Interspeech 2020, 4646-4650, DOI: 10.21437/Interspeech.2020-1977.


@inproceedings{O’Connor2020,
  author={Matt O’Connor and W. Bastiaan Kleijn},
  title={{Distributed Summation Privacy for Speech Enhancement}},
  year=2020,
  booktitle={Proc. Interspeech 2020},
  pages={4646--4650},
  doi={10.21437/Interspeech.2020-1977},
  url={http://dx.doi.org/10.21437/Interspeech.2020-1977}
}