19 Nov
1999
Sphinx
Speech Group, CMU-SCS (rkm@cs.cmu.edu)
Sphinx-3 Search Algorithm
initialize start
state of <S> with path-score = 1;
for each frame of
input speech {
evaluate all active HMMs; find best
path-score, pruning thresholds;
for each active HMM {
if above pruning threshold
{
activate HMM for next
frame;
transition to and
activate successor HMM within word, if any
if word-final HMM and
above word-pruning threshold
record
word-exit in BP table;
}
}
transition from words exited into
initial state of entire lexicon (using the
LM), and activate HMMs
entered;
}
find final </S> BP
table entry and back-trace through table to retrieve result;