I developed a simple android app (on a smartwatch) where user can simply start and stop specific activities by simply tap on the screen (and a visual feedback assess the start or stop). I want to enhance this app by adding voice recognition capability thanks to the pocketsphinx tools.
I read the CMUSphinx tutorials, and I am still not totally sure about which langage model I should use.
App description :
It’s kind of command and control action. When a specific activity is selected (manually) on the app, the user shall be able to start the activity by voice by saying a simple keyword as « Start Activity ». Any other words can be said before or after that command, and it should not affect the app behavior. Once activity started, the user should be able to stop it whenever he wants by saying « Stop activity ». One more time, everything can be said between those to actions.
My questions : What kind of langage model should I use ?
-Should I simple use a KeyWord spotting mode for « Start activity » and « stop activity » that is always activated. In that case, it’s still necessary to have a full vocabulary dictionary, or is it better for accuracy to make a small one with only the keywords ?
-Should I use grammar which seems to be appropriate with command and control app ? But when a grammar search is activated, I understood that only the specific grammar words should be pronounced ? This is not suitable for the app goal.
Thanks a lot in advance
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
In that case, it’s still necessary to have a full vocabulary dictionary, or is it better for accuracy to make a small one with only the keywords ?
Dictionary size does not affect accuracy, it only saves memory.
-Should I use grammar which seems to be appropriate with command and control app ? But when a grammar search is activated, I understood that only the specific grammar words should be pronounced ? This is not suitable for the app goal.
I think you answered your own question yourself.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
By any chance (I look every where but didn't find), is somewhere a JavaDoc available for pocketsphinx android ?
Or a kind of doxyfile which describes the available functions that are not used in the demo (to manipulate all the more "complexe" parameters of a Search etc... ?
Would be so helpful !
Have a good day !
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi,
I developed a simple android app (on a smartwatch) where user can simply start and stop specific activities by simply tap on the screen (and a visual feedback assess the start or stop). I want to enhance this app by adding voice recognition capability thanks to the pocketsphinx tools.
I read the CMUSphinx tutorials, and I am still not totally sure about which langage model I should use.
App description :
It’s kind of command and control action. When a specific activity is selected (manually) on the app, the user shall be able to start the activity by voice by saying a simple keyword as « Start Activity ». Any other words can be said before or after that command, and it should not affect the app behavior. Once activity started, the user should be able to stop it whenever he wants by saying « Stop activity ». One more time, everything can be said between those to actions.
My questions : What kind of langage model should I use ?
-Should I simple use a KeyWord spotting mode for « Start activity » and « stop activity » that is always activated. In that case, it’s still necessary to have a full vocabulary dictionary, or is it better for accuracy to make a small one with only the keywords ?
-Should I use grammar which seems to be appropriate with command and control app ? But when a grammar search is activated, I understood that only the specific grammar words should be pronounced ? This is not suitable for the app goal.
Thanks a lot in advance
Dictionary size does not affect accuracy, it only saves memory.
I think you answered your own question yourself.
Thanks for your answer Sir !
By any chance (I look every where but didn't find), is somewhere a JavaDoc available for pocketsphinx android ?
Or a kind of doxyfile which describes the available functions that are not used in the demo (to manipulate all the more "complexe" parameters of a Search etc... ?
Would be so helpful !
Have a good day !
https://cmusphinx.github.io/doc/pocketsphinx/pocketsphinx_8h.html