Segment-Based Acoustic Models with Multi-level Search Algorithms for 
Continuous Speech Recognition 
Mari Ostendorf J. Robin Rohlicek 
Boston University BBN Inc. 
Objective: 
The goal of this project is to develop improved acoustic models for speaker-independent recog- 
nition of continuous speech, together with efficient search algorithms appropriate for use with 
these models. The current work on acoustic modelling is focussed on stochastic, segment-based 
models that capture the time correlation of a sequence of observations (feature vectors) that 
correspond to a phoneme. Since the use of segment models is computationally complex, we 
will also investigate multi-level, iterative algorithms to achieve a more efficient search. Fur- 
thermore, these algorithms will provide a formalism for incorporating higher-order information. 
This research is jointly sponsored by DARPA and NSF. 
Summary of Accomplishments: 
• Investigated the effect on recognition performance of different parameters of the segment 
model, particularly as a function of training. 
• Developed a new approach to modeling time correlation in the context of the segment 
model. 
Developed new fast approaches to phone recognition, motivated by algorithms from image 
segmentation, which offers significant computational advantages over dynamic program- 
ming. 
• Developed a reformulation of hidden Markov models which allows for more general map- 
pings from the acoustic features to state likelihoods. 
• Achieved recognition results on the TIM1T database using context-independent models 
which are comparable to those reported by others using context-dependent models. 
Plans: 
• Investigate dynamical system models for representing time correlation and context-dependence 
in the segment model. 
• Extend current results to use segmental features and context-dependent models. 
• Investigate mechanisms for integrating segment algorithms with the BBN Byblos recog- 
nition system. 
• Investigate global constraints and features using multi-level algorithms. 
409 
