Signal Processing - November 2016 - 57

The increasing automation of driving may allow cars to
with respect to form factor, and the speech channel is little
contain larger screens. Virtual reality or mixed reality could
utilized in single- occupant vehicles.
make use of this screen space to present content, possibly as
At all points on the vehicle autonomy spectrum, drivers
a three-dimensional (3-D) virtual world. CID systems may
and passengers can avail themselves of Internet connectivplay a special role in navigating such 3-D worlds while drivity to rich content and services only recently accessible ining; for example, one may, with a speech request, switch the
vehicle. Time in the car can be used to entertain oneself or
display to a place not shown on a screen and preview
complete useful tasks, to the extent that these activities do
its surroundings.
not interfere with driving responsibilities. Audio channel is
Recognizing that CID technology can both increase drivpreferable to visual for these activities because reading in
ing safety and improve user experience allows one to see new
cars causes motion sickness for many people and because
useful opportunities for in-car speech systems. These, howvehicle vibration caused by road conditions interferes with
ever, require more conversational intelligence than is curreading but listening is much less affected. Therefore, CID
rently available to ensure that drivers' and passengers' voice
systems will remain a desirable interface channel for content
interactions with systems are natural and not cognitively
and service consumption.
demanding or distracting.
When drivers have greater responsibility for controlling their vehicles, CID systems hold the potential to
reduce distraction and increase safety by, for example,
CID systems in the context of the Internet of Things
communicating information about vehicle health and
Today, a single car can contain more than 100 sensors, suproad conditions in natural language instead of with crypporting various vehicle functionalities and detecting occupant
tic warning lights or not at all. When autonomous cars take
states. As vehicle technology advances, increasing numbers
over greater responsibility for controland types of sensors will appear in cars.
ling vehicles, the drivers' trust or sense
At all points on the vehicle Some of these will improve vehicle autoof a safe ride need to be built up over
mation and safety, some will sense the
autonomy spectrum,
time. This trust-building process for the
external environment, for example, meadrivers and passengers
adoption of autonomous vehicles can be
sure air quality, and others will sense occucan avail themselves
facilitated by the CID systems through
pants' physical states and actions, such as
communicating information about vehiposture or alertness. Such sensors can proof Internet connectivity
cle controlling capability together with
vide much of the information needed for
to rich content and
vehicle health and road conditions.
improving the conversational intelligence
services only recently
Further along the autonomy spectrum,
of CID systems.
accessible in-vehicle.
where smarter cars and intelligent trafWith the increase of connectivity in
fic systems provide ever greater driver
the Internet of Things (IoT), devices
assistance, the temptation for drivers to pay less attention
and sensors will increasingly become more diverse
to driving and more to unrelated content and services will
with much richer information-exchange functionalities,
naturally grow as the primary driving tasks diminish. The
expressing additional device status, features, operating
potential for boredom increases drivers' risk of becoming
instructions, or maintenance needs, etc. With new form
inattentive and occupying themselves with nondriving activfactors of devices and sensors, the success of touch screens
ities. Smarter cars equipped with smarter speech systems
on smart phones for information exchange may not be eascan reduce this risk of distraction by keeping drivers engaged
ily repeated in cars where physical spaces are limited or
to an appropriate extent, and preparing them to take over
design flexibility is required. As the screen size decreasgreater vehicle control as necessary. For instance, such cars
es, speech will become a more preferred in formation excould explain aloud their automated responses to significhange modality.
cant changes in driving circumstances. When approaching
The IoTs will also bring in much more content and many
a complex intersection, encountering difficult vehicle or
more services for drivers to access. One may receive a scepedestrian traffic, or upon detecting an unexpected road
nic spot description or hotel vacancy advertisement along a
closure, cars can announce pertinent information preparing
highway not through traditional billboards but rather from
the driver to handle the situations.
the Internet in real time. The general IPAs may intend to
When fully autonomous driving becomes available, transisupport people for these needs. However, given the typically
tions from autonomous-driving mode to human-driver mode
fast-changing environment involved in driving, the in-vehiand vice versa will need to be very intuitive and natural
cle use cases add much more dynamic and context-sensitive
without any additional training. When the vehicle requires a
requirements for such assistance. In-vehicle CID systems
driver to take over the control, the most natural way for drivhave the potential in using in-vehicle sensors, such as gasers will be for the alert to come via speech request. Liketank level, to find better solutions for the drivers. One will
wise, allowing drivers to request the vehicle to take over
expect a tight integration of the general IPAs and embedded
control using speech commands will be natural and conveCID systems to offer the drivers synergized benefits from
nient for drivers.
both systems.
IEEE SIgnal ProcESSIng MagazInE

November 2016

Table of Contents for the Digital Edition of Signal Processing - November 2016