Acoustic-Phonetics Based Speech Recognition 
Victor W. Zue 
Spoken Language Systems Group 
Laboratory for Computer Science 
Massachusetts Institute of Technology 
OBJECTIVE: 
The objective of this project is to develop a robust and high-performance speech recognitiotl 
system using a segment-based approach to phonetic recognition. The recognition system 
will eventually be integrated with natural language processing to achieve spoken lallguagc 
understanding. 
SUMMA R Y OF ACCOMPLISHMENTS: 
Developed a phonetic recognition front-end and achieved 77% and 71% classiilcatiou 
accuracy under speaker-dependent and -independent conditions, respectively, using a 
set of 38 context-independent models. 
Collaborated with researchers at SRI in the development of the MISTRI system, mak- 
ing explicit use of acoustic-phonetic and phonological knowledge. 
Developed the SUMMIT speech recognition system that incorporates auditory mod- 
elling and explicit segmentation, and achieved a speaker-independent accuracy of 87% 
on the DARPA 1000-word Resource Management task using 75 phoneme models. 
Developed probabilistic natural language system, TINA, and achieved a test-set cov- 
erage of 78% with perplexity of 42 for the Resource Management task. 
• Transcribed all 6300 sentences for the TIMIT database. 
Developed a set of research tools for the DARPA speech research community in ot'dcr 
to facilitate data collection, parameter computation, statistical analysis, and speech 
synthesis. 
PLANS: 
Improve the speech recognition performance by incorporating context-dependency ia 
phoneme modelling. 
Integrate TINA into SUMMIT in order to develop spoken language understanding 
capabilities. 
Develop a back-end on the task of a Knowledgeable Navigator, and integrate it with 
the spoken language system. 
Begin hardware development, such that the system will soon be able to execute in near 
real-time. 
167 
