I found that there is trade-off between reliability of acoustic score
(normalized by number of frames) of a decoded word and the total number of
words in dictionary. As I go on reducing the number of words in dictionary and
LM the acoustic scores become more and more reliable to say whether the
decoded word can be accepted as correct or wrong. I could not understand why
this happens. Can any one explain ?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
My result was using sphinx3.0.6 . What do you mean by best score ? Best score
of what ? And even if it normalizes, why should it lead to this inverse
relation between reliability of acoustic scores and size of dic ?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I found that there is trade-off between reliability of acoustic score
(normalized by number of frames) of a decoded word and the total number of
words in dictionary. As I go on reducing the number of words in dictionary and
LM the acoustic scores become more and more reliable to say whether the
decoded word can be accepted as correct or wrong. I could not understand why
this happens. Can any one explain ?
It depends on the decoder. Pocketsphinx for example normalize acoustic score
each frame by best score.
My result was using sphinx3.0.6 . What do you mean by best score ? Best score
of what ? And even if it normalizes, why should it lead to this inverse
relation between reliability of acoustic scores and size of dic ?