I train an acoustic model for a system with a very small number of words.
For example 10 numbers (in italian language):
ONE
TWO
THREE
FOUR
FIVE
....
I create a little database with 55 speakers, each speaker repeat the words 5 times.
I test the CI model created and i have obtained good results in terms of WER.
Now i try my model on a utterances (wav signals) that contain words not in a dictionary.
The results as i predicted are not satisfatory.
In particular using my trained model all words not contained in dictionary are tagged as word in vocabulary.
Can i use any score to find this false positive, or i must create a new model?
Thk
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi -
I train an acoustic model for a system with a very small number of words.
For example 10 numbers (in italian language):
ONE
TWO
THREE
FOUR
FIVE
....
I create a little database with 55 speakers, each speaker repeat the words 5 times.
I test the CI model created and i have obtained good results in terms of WER.
Now i try my model on a utterances (wav signals) that contain words not in a dictionary.
The results as i predicted are not satisfatory.
In particular using my trained model all words not contained in dictionary are tagged as word in vocabulary.
Can i use any score to find this false positive, or i must create a new model?
Thk
I am confused by your question.
If you want to recognize other words, just add them to the dictionary. Its unlikely to be very successful due the size of your acoustic model.
I am not sure what you mean by 'score to find this false positive , or i must create a new model?'