Instrumentation & Measurement Magazine 24-6 - 17

interacted with virtual
assistants [25]; achieved
79% accuracy in predicting
future car accidents
(within 1.5 years) based on
pauses, filled pauses, and
pronoun use in conversations
with virtual assistants
[26]; and found that people
with MCI pause more
often and for longer, and
they produce fewer words
and shorter speech chunks
when interacting with virtual
assistants than do
cognitively healthy users
[27].
In addition to the existing
commercial offerings,
another type of digital asFig.
3. Some of the advantages and challenges associated with smart home speech-based measurement of cognitive
status.
sistant may soon be available: personal care robots. Often
proposed as an assistive technology to allow older adults to
live in their homes longer, voice-controlled robots could also
record and analyze the user's voice for signs of cognitive decline.
In one study, a conversational robot was found to be
generally well-liked and engaging by participants with AD,
and a number of the features extracted from the dialogues distinguished
between persons with mild, moderate, and severe
dementia [28].
Ambient embedded sensors: Perhaps the most comprehensive
speech sensing methodology is to embed microphones
throughout the smart home and record everything that occurs.
Embedded microphone arrays have been previously
proposed to help detect falls and to recognize cries for help.
The SWEET-HOME project in France focused intensively on
bringing audio technology to smart homes for older people
and built a home lab with 150 sensors and actuators, including
multiple omnidirectional microphones. They found that audio
recording was much more acceptable to older users than video
recording [29]. The focus of that and other audio-based smart
home projects has typically been on providing assistive technologies,
not assessing cognitive state, though there is clearly
potential for further research. For example, it has been proposed
that ambient assistive living technologies for persons
with dementia could additionally report some behavioral metrics
to the care team, including instances of repetitive speech
[30]. However, privacy preservation is paramount in the extensive
recording scenarios discussed in this section.
Challenges
Thus far, we have focused on the benefits of smart home
speech recording for cognitive monitoring: it can be continuous,
unobtrusive, real-time, longitudinal, inclusive to those
with mobility issues or who live in remote areas, can avoid unreliable
self-reported data, and is potentially more sensitive to
September 2021
decline than traditional cognitive tests [2], [13], [22], [31]. However,
there are a number of serious challenges which also must
be addressed, both on the technical side and from the ethical
perspective. These challenges are summarized in Fig. 3 and
further detailed in the following sections.
Technical Challenges
There are technical challenges associated with continuously
recording, transmitting, and analyzing audio data that may
not arise or be addressed in lab-based feasibility studies. The
choice of the recording device itself is obviously a critical decision.
As Shirmohammadi et al. [32] point out in their review
on biomedical instrumentation and measurement, the primary
question in home-based biomarker measurement is:
" How can we safeguard that the clinical quality of measurements
is achieved using the low-cost measurement equipment
in a home environment? " For microphones, the sensitivity, frequency
response, and sampling rate will all affect the quality
of the audio data.
For recording the data, sensor location relative to the
speaker is also important. If using a mobile or wearable device,
is the microphone obscured by clothing? If using a virtual
assistant or embedded microphones, the distance between
speaker and sensor will be constantly changing, and in the case
of a robotic assistant, the acoustic properties of the environment
will also vary, leading to complications when processing
the data.
Once the data is recorded, it must be transmitted, typically
to an encrypted server for processing. Some on-device
processing may be possible, but in most cases processing and
storage will occur on an external server. One advantage to
home-based systems is that they can be connected to wired internet
and a continuous power source, avoiding issues due
to lost Wi-Fi connection or drained batteries. However, any
time data is transmitted there is some vulnerability to security
threats. Issues of data security, integrity, confidentiality, and
IEEE Instrumentation & Measurement Magazine
17

Instrumentation & Measurement Magazine 24-6

Table of Contents for the Digital Edition of Instrumentation & Measurement Magazine 24-6