Anna Leschanowsky(Aalto University), Sneha Das(Aalto University), Tom Bäckström(Aalto University) and Pablo Pérez Zarazaga(Aalto University)
Abstract:
Voice based devices and virtual assistants are widely integrated
into our daily life, but the growing popularity has also raised
concerns about data privacy in processing and storage. While
improvements in technology and data protection regulations
have been made to provide users a more secure experience, the
concept of privacy continues to be subject to enormous challenges.
We can observe that people intuitively adjust their way
of talking in a human-to-human conversation, an intuition that
devices could benefit from to increase their level of privacy. In
order to enable devices to quantify privacy in an acoustic scenario,
this paper focuses on how people perceive privacy with
respect to environmental noise. We measured privacy scores on
a crowdsourcing platform with a paired comparison listening
test and obtained reliable and consistent results. Our measurements
show that the experience of privacy varies depending on
the acoustic features of the ambient noise. Furthermore, multiple
probabilistic choice models were fitted to the data to obtain
a meaningful ordering of noise scenarios conveying listeners’
preferences. A preference tree model was found to fit best, indicating
that subjects change their decision strategy depending
on the scenarios under test.