Sine-Wave Speech and Privacy-Preserving Depression Detection
Lists
Only you see this
Abstract
In this work we show that certain characteristics of speech important for depression detection have future potential to allow privacy preservation. Specifically, we are interested in detecting depression from a limited set of acoustic properties of speech, such as, for example, the lower speech formants (F1, F2, F3). Although the full set of formants is generally sufficient to recognize most of the semantic content of the speech, this does not appear to be the case for a single formant. This gives rise to the basic idea we present here, which is (a) to use sine-wave speech for detecting depression based on a strict subset of the formants, together with (b) an initial exploration of the privacy-preserving characteristics of this sine-wave speech. Privacy is an important concern in the application of machine learning techniques to healthcare applications, and our work points towards the future possibility of, for example, sharing anonymized versions of sensitive data (in fact, versions stripped of all semantic content) that still allow some analyses with respect to presence of depression.
BibTeX
@inproceedings{dumpala2021sine,
author = {Sri Harsha Dumpala and Rudolf Uher and Stan Matwin and Michael Kiefte and Sageev Oore},
title = {{Sine-Wave Speech and Privacy-Preserving Depression Detection}},
booktitle = {SMM},
year = {2021},
url = {https://openreview.net/forum?id=TzFZ8vk87M},
}Loading related content...