Hello,
How is the "score" in pocketsphinx's function
ps_nbest_hyp(nbest, &score)
calculated?
I'm asking this because I noticed that this score is different from the score that the function
ps_get_hyp(ps, &score, &uttid)
gives.
E.g. the unit test
test_ps_nbest
gives:
BESTPATH (goforward): go forward and users (-29691)
.. but the nbest score of the same hypothesis is very different:
NBEST 1: go forward and users (-38624)
I tried to dig in the code in src/libpocketsphinx/ps_lattice.c but the understanding curve is very steep, hence the question, sorry about that.
Wish I know that too. It might be a bug.
I would do experiments on simple lattices first which you can load from files. Then this can be extended to a real speech lattices.
Log in to post a comment.
Hello,
How is the "score" in pocketsphinx's function
calculated?
I'm asking this because I noticed that this score is different from the score
that the function
gives.
E.g. the unit test
gives:
.. but the nbest score of the same hypothesis is very different:
I tried to dig in the code in src/libpocketsphinx/ps_lattice.c but the
understanding curve is very steep, hence the question, sorry about that.
Wish I know that too. It might be a bug.
I would do experiments on simple lattices first which you can load from files.
Then this can be extended to a real speech lattices.