tech,21H921036,bq 
% reduction in error . We discuss
<term>

maximum a posteriori estimation

</term>
of
<term>
continuous density hidden

#19054
We discussmaximum a posteriori estimation of continuous density hidden Markov models (CDHMM). 
tech,82H921036,bq 
reestimation algorithms
</term>
, namely the
<term>

forwardbackward algorithm

</term>
and the
<term>
segmental kmeans algorithm

#19076
The classical MLE reestimation algorithms, namely theforwardbackward algorithm and the segmental kmeans algorithm, are expanded and reestimation formulas are given for HMM with Gaussian mixture observation densities. 
tech,63H921036,bq 
. Because of its adaptive nature ,
<term>

Bayesian learning

</term>
serves as a unified approach for

#19105
Because of its adaptive nature,Bayesian learning serves as a unified approach for the following four speech recognition applications, namely parameter smoothing, speaker adaptation, speaker group modeling and corrective training. 
tech,173H921036,bq 
unified approach for the following four
<term>

speech recognition

</term>
applications , namely
<term>
parameter

#19116
Because of its adaptive nature, Bayesian learning serves as a unified approach for the following fourspeech recognition applications, namely parameter smoothing, speaker adaptation, speaker group modeling and corrective training. 
tech,122H921036,bq 
forwardbackward algorithm
</term>
and the
<term>

segmental kmeans algorithm

</term>
, are expanded and
<term>
reestimation

#19080
The classical MLE reestimation algorithms, namely the forwardbackward algorithm and thesegmental kmeans algorithm, are expanded and reestimation formulas are given for HMM with Gaussian mixture observation densities. 
other,192H921036,bq 
algorithm
</term>
, are expanded and
<term>

reestimation formulas

</term>
are given for
<term>
HMM with Gaussian

#19087
The classical MLE reestimation algorithms, namely the forwardbackward algorithm and the segmental kmeans algorithm, are expanded andreestimation formulas are given for HMM with Gaussian mixture observation densities. 
tech,22H921036,bq 
models ( CDHMM )
</term>
. The classical
<term>

MLE reestimation algorithms

</term>
, namely the
<term>
forwardbackward

#19070
The classicalMLE reestimation algorithms, namely the forwardbackward algorithm and the segmental kmeans algorithm, are expanded and reestimation formulas are given for HMM with Gaussian mixture observation densities. 
tech,253H921036,bq 
namely
<term>
parameter smoothing
</term>
,
<term>

speaker adaptation

</term>
,
<term>
speaker group modeling
</term>

#19124
Because of its adaptive nature, Bayesian learning serves as a unified approach for the following four speech recognition applications, namely parameter smoothing,speaker adaptation, speaker group modeling and corrective training. 
tech,223H921036,bq 
recognition
</term>
applications , namely
<term>

parameter smoothing

</term>
,
<term>
speaker adaptation
</term>
,

#19121
Because of its adaptive nature, Bayesian learning serves as a unified approach for the following four speech recognition applications, namelyparameter smoothing, speaker adaptation, speaker group modeling and corrective training. 
tech,323H921036,bq 
<term>
speaker group modeling
</term>
and
<term>

corrective training

</term>
. New experimental results on all

#19131
Because of its adaptive nature, Bayesian learning serves as a unified approach for the following four speech recognition applications, namely parameter smoothing, speaker adaptation, speaker group modeling andcorrective training. 
model,71H921036,bq 
maximum a posteriori estimation
</term>
of
<term>

continuous density hidden Markov models ( CDHMM )

</term>
. The classical
<term>
MLE reestimation

#19059
We discuss maximum a posteriori estimation ofcontinuous density hidden Markov models ( CDHMM ). 
model,242H921036,bq 
reestimation formulas
</term>
are given for
<term>

HMM with Gaussian mixture observation densities

</term>
. Because of its adaptive nature

#19092
The classical MLE reestimation algorithms, namely the forwardbackward algorithm and the segmental kmeans algorithm, are expanded and reestimation formulas are given forHMM with Gaussian mixture observation densities. 
tech,283H921036,bq 
</term>
,
<term>
speaker adaptation
</term>
,
<term>

speaker group modeling

</term>
and
<term>
corrective training
</term>

#19127
Because of its adaptive nature, Bayesian learning serves as a unified approach for the following four speech recognition applications, namely parameter smoothing, speaker adaptation,speaker group modeling and corrective training. 