19 Nov
1999
Sphinx
Speech Group, CMU-SCS (rkm@cs.cmu.edu)
Word Segmentation Problem
Q: Which
transition wins?
·Flat
lexicon (separate model per word):
·At
A: word ninety entered with
LM score P (ninety
| ninety)
·At
B: word ninety entered with
P (ninety
| nineteen)
·Since
the latter is much better, it prevails over the former
·Result:
correct recognition, and segmentation for ninety
nineteen
ninety
ninety
ninety
nineteen
ninety
ninety
correct segmentation for ninety
incorrect segmentation for ninety
Assume P (ninety | nineteen) > P (ninety | ninety).
Decision points: A B C