Gisting Continuous Speech 
J. R. Rohlicek, Principal Investigator 
BBN Systems and Technologies 
70 Fawcett St. 
Cambridge, MA 02138 
PROJECT GOALS 
The objective of this work is automatic, real-time "gisting" 
of voice: traffic for updating of information in databases, 
for producing timely reports, and for prompt notification of 
events of interest. Specifically, the goal is to build a proto- 
type, real-time system capable of processing radio commu- 
nication between air traffic controllers and pilots; identifying 
dialogs and extracting their "gist" (e.g., identifying flights, 
determining whether they are landing or taking off), and 
producing a continuous output stream with that information. 
The approach is intended to be general and applicable to 
other domains. 
The system is built upon state-of-the-art techniques in speech 
recognition, speaker identification, natural language anal- 
ysis, and topic statistical classification. These techniques 
have been extended where necessary to address specific as- 
pects of the gisting problem. Because various sources of 
information must be combined, the system design features a 
itigh degree of interaction between the natural language and 
domain-knowledge components and the speech processing 
components. 
RECENT RESULTS 
We have made additions and modifications to our prototype 
system \[1\]. The primary goal of the effort was to achieve 
real-time performance. This involved both system architec- 
tural and algorithmic modifications described fully in \[2\]. 
A prototype system has been evaluated using approximately 
14 hours of data recorded at Logan airport. Performance 
was measured on appro~nately four hours of data held out 
for final evaluation. On that data, the system achieved ap- 
proximately 88% recall and 82% precision for detection of 
controller-pilot dialogs. Also, of the flights correctly de- 
tected, the fight identification was correctly extracted 59% 
of the time. 
A real-time prototype system has been constructed. The 
system builds on a flexible software system developed as 
part of this effort. The system allows multiple processes to 
be coordinated across multiple hosts and provides facilities 
for efficient stream connections between modules as well as 
flexible message-based communication between modules. 
PLANS FOR THE COMING YEAR 
The remainder of the effort will focus on completion of 
the prototype system and on system testing. System testing 
includes investigation of the sensitivity of overall system 
performance to the performance of various component. 
REFERENCES 
\[1\] J. R. Rohlicek, et al, "Gisling conversational speech," 
Proceedings of the International Conference on Acoustics, 
Speech and Signal Processing, Mar 1992, pp. II-113. 
\[2\] L. Denenberg, et. al., "Gisting conversational speech in 
real time," to appear in Proceedings of the International Con- 
ference on Acoustics, Speech and Signal Processing, April 
1993. 
384 
