Advances in vision, speech, and sensor technologies in the recent
years have opened up a number of new and exciting avenues for
ubiquitous computing research. One emerging application in ubicomp
is Smart Spaces or Intelligent environments, the goal of which is
to augment our conventional living and working spaces with computing
and inference power in a way that would enable them to be more aware or
cognizant of changes, particularly in relation to the activities and
behaviors of their occupants. Based on this knowledge, the physical world
can learn to adapt to better accommodate to people's diverse needs.
Smart Space is not a single application -- it is a platform on which a
vast number of applications can be built and tested. The canonical example
is automatic adjustment of room conditions, such as for lighting, temperature,
or stereo volume, based on detection of user identity and prior knowledge of
their preferences. Smart Space technology has many applications in the health
domain. One such example is sensing and monitoring the well-being of its occupants
(espcially for the elderly) and alert relevant personnel in case of emergencies.
In many ways, Smart Space can be likened to the house butler, albiet it is invisible
to the occupants.
At the BiD, we are in the process of creating a smart work space that
ultilizes advanced speech, vision, and sensing technologies. One
critical open issue to the success of this class of applications is
their ability to predict the nature of the events that they "observe"
with a high degree of accuracy. We approach this issue from a HCI standpoint
by leveraging contextual information into the event inference system.
Context which includes information regarding people's identity and the
activities in which they are engaged shall play a critical role in making
sensible and accurate prediction about external events.
Speaker localization and identification are two important smart room technologies
that would allow us to gather information to construct location-based contexts. We
are investigating the use of large microphone arrays for estimating and tracking
the spatial locations of active sound sources in a lab environment.
A set of 36 dynamic, unidirectional microphones are arranged into a 6 by 6 grid to
capture speech signals. The data are channeled by 828mkII firewire audio interfaces
to the processing subsystem.
The position of active sound sources are computed by running a combination of
digital signal processing techniques and numerical algorithms on the raw speech
signals. The movement of sound sources and their relative intensities are visualized
using a Java applet.
We plan to extend the system with the ability to distinguish different "speakers"
from the intermingled speech data. This would allow us to construct a clear picture
on which areas of the room people spend most of their time in. This information could
then be used to inform design to better accomodate the needs of users.
