Hello ,
i have a database with 200 arabic words . 25 person . 5 itterations for every word .
i would now to make an application speech to text using C++ ( QT creator ) .
Wich model i must use in training , Semi or Cont ?
i have do a trainig with semi model and it takes more than 3.5 hours and i had get an accuracy 78% .
i have do a training with cont model and it takes 30 minutes and i had get an accuracy 80% .
I'm sure that their is a problem in cont model ?
how can i solve that ?
and what is the best configuration that i must do ?
thanks a lot .
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
thanks sir
can i use Semi model with C++ ?
i had use continuos model in my c++ application and it get many recognition errors .
i have a button that choose a file then a recognition button .
The recognition using the application is 40% i guess . and using the same test base that i had use in Sphinxtrain .
So , why ?
in Sphinxtrain 80% and in the application 40% ? .
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
how can i use Semi model with C++ ?
because when i put the Semi Parametrs PATH it does not work .
but with the cont parametrs path it work
-cmninit 40,3,-1 : What it mean this line ?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I had success in doing Semi model with my c++ application .
please can you tell me what this line mean
-cmninit 40,3,-1
and what 40 , 3 , -1 mean
thanks
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hello ,
i have a database with 200 arabic words . 25 person . 5 itterations for every word .
i would now to make an application speech to text using C++ ( QT creator ) .
Wich model i must use in training , Semi or Cont ?
i have do a trainig with semi model and it takes more than 3.5 hours and i had get an accuracy 78% .
i have do a training with cont model and it takes 30 minutes and i had get an accuracy 80% .
I'm sure that their is a problem in cont model ?
how can i solve that ?
and what is the best configuration that i must do ?
thanks a lot .
There is no problem with continuos model if speed is good for you.
To get better accuracy you need to collect more data.
Optimal parameters require tuning, you can't guess them without access to the database.
thanks sir
can i use Semi model with C++ ?
i had use continuos model in my c++ application and it get many recognition errors .
i have a button that choose a file then a recognition button .
The recognition using the application is 40% i guess . and using the same test base that i had use in Sphinxtrain .
So , why ?
in Sphinxtrain 80% and in the application 40% ? .
Yes
You need to add a line
to the file
feat.paramsinside the modelhow can i use Semi model with C++ ?
because when i put the Semi Parametrs PATH it does not work .
but with the cont parametrs path it work
-cmninit 40,3,-1 : What it mean this line ?
-cmninit 40,3,-1
when i had add it to the feat.params
the application does not recognize any word and the hyp variable still NULL
Well, to get help you can do the following:
1) Provide acoustic model training folder
2) Provide recordings you are trying to recognize and the recognition logs.
Thanks
here is the Accoustic cont model training + dic + lm.DMP + wav recordings ( 16bit , mono , 16khz ) + result file .
I had success in doing Semi model with my c++ application .
please can you tell me what this line mean
-cmninit 40,3,-1
and what 40 , 3 , -1 mean
thanks