19 Nov
1999
Sphinx
Speech Group, CMU-SCS (rkm@cs.cmu.edu)
·Incremental
LM score accumulation; e.g. (bigram LM):
·Large
computation and memory requirements
·Overhead
for dynamic lextree creation/destruction
LM Integration: Lextree Replication
A
B
S = set of words reachable from x
x
W e S
LM probability entering into x = Sum [P(W | B)]
B
Lextree
A
Lextree
B
Lextree
C