How does Pocketsphinx handle out of vocabulary words? It seems that they are
forced to fit a word that is in the vocabulary. Is it assumed that the user
will only use the words in the vocabulary? The purpose of my application is
command and control so there is no reason to need to recognize out of
vocabulary words, just that they are not in the vocabulary.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
How does Pocketsphinx handle out of vocabulary words?
I don' t think "handle" is applicable to the speech recognition engine.
It seems that they are forced to fit a word that is in the vocabulary. Is it
assumed that the user will only use the words in the vocabulary?
It's not forced and not assumed. The algorithm is just built this way to
recognize only words in the grammar
The purpose of my application is command and control so there is no reason
to need to recognize out of vocabulary words, just that they are not in the
vocabulary.
There is no need but you still should make your grammar as complete as
possible. That helps to with result accuracy. Good practice is usually to
include all possible words into the grammar including words user might say.
Later you need to have to analyse recognition result to extract the actual
command.
Next, you need to include garbage loop into your grammar like single-phone
words with lower probability to recognize them in order to handle words that
remain uncovered.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Anonymous
-
2010-08-16
Can you recommend a resource for how to implement a garbage loop?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
How does Pocketsphinx handle out of vocabulary words? It seems that they are
forced to fit a word that is in the vocabulary. Is it assumed that the user
will only use the words in the vocabulary? The purpose of my application is
command and control so there is no reason to need to recognize out of
vocabulary words, just that they are not in the vocabulary.
I don' t think "handle" is applicable to the speech recognition engine.
It's not forced and not assumed. The algorithm is just built this way to
recognize only words in the grammar
There is no need but you still should make your grammar as complete as
possible. That helps to with result accuracy. Good practice is usually to
include all possible words into the grammar including words user might say.
Later you need to have to analyse recognition result to extract the actual
command.
Next, you need to include garbage loop into your grammar like single-phone
words with lower probability to recognize them in order to handle words that
remain uncovered.
Can you recommend a resource for how to implement a garbage loop?