I am using pocketsphinx and created my own lm and dict files and it's working
great. I haven't spent a lot of time decifering all the stats that continuos
gives out, but i am hoping someone can answer a simple question. Is there a
stat that will give me the level of certainty that the word was recognized ?
It is very accurate but i have noticed that continuos sometimes will tell me
"not found in last frame, using ____ instead" and other times it seems to
recognize the same word perfectly and does not display that. In both cases the
correct word is printed on the screen, but i'd like to know if it's having
trouble recognizing it against the grammar file.
Thanks in advance
John
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Is there a stat that will give me the level of certainty that the word was
recognized ?
There are ps_get_prob and ps_seg_prob API calls
t is very accurate but i have noticed that continuos sometimes will tell me
"not found in last frame, using ____ instead" and other times it seems to
recognize the same word perfectly and does not display that.
It's expected behaviour, nothing to worry about.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I tried both ps_get prob and ps_seg prob and both appeared to give me nonsense
in pocketsphinx_continuous in the recognize_from_microphone() function after
ps_get_hyp()
i tried:
prob = ps_get_prob(ps, &uttid);
and the majority of the time prob returns 0 and other times it will return a
large negative number. I'll say the same words over and over again that i know
are not in my grammar file and it does not appear to give me useful
information.
Any ideas?
(i also tried ps_seg_prob and didn't see any consistent results from the
accoustic or language score (again just large negative numbers)
thanks again for the help
John
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
If you think that results are not good enough you can always provide a
testcase for the problem you see. Without testcase it's hard to suggest you
anything.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I am using pocketsphinx and created my own lm and dict files and it's working
great. I haven't spent a lot of time decifering all the stats that continuos
gives out, but i am hoping someone can answer a simple question. Is there a
stat that will give me the level of certainty that the word was recognized ?
It is very accurate but i have noticed that continuos sometimes will tell me
"not found in last frame, using ____ instead" and other times it seems to
recognize the same word perfectly and does not display that. In both cases the
correct word is printed on the screen, but i'd like to know if it's having
trouble recognizing it against the grammar file.
Thanks in advance
John
There are ps_get_prob and ps_seg_prob API calls
It's expected behaviour, nothing to worry about.
I tried both ps_get prob and ps_seg prob and both appeared to give me nonsense
in pocketsphinx_continuous in the recognize_from_microphone() function after
ps_get_hyp()
i tried:
prob = ps_get_prob(ps, &uttid);
and the majority of the time prob returns 0 and other times it will return a
large negative number. I'll say the same words over and over again that i know
are not in my grammar file and it does not appear to give me useful
information.
Any ideas?
(i also tried ps_seg_prob and didn't see any consistent results from the
accoustic or language score (again just large negative numbers)
thanks again for the help
John
Hello
If you think that results are not good enough you can always provide a
testcase for the problem you see. Without testcase it's hard to suggest you
anything.
whats the easiest way to submit a test case to you?
thank you
Upload an archive to a public file sharing resource and give here a link.