Hello,
I want to change my dictionary and my JSGF grammar file without reloading the acoustic model.
However it seems that I get very bad results if I do this without reinitializing the whole decoder. Indeed, if I reload everything I get an accuracy higher than 90% but if I only update the dictionary using ps_add_word(), nd change the JSGF using fsg_set_remove_byname(), jsgf_parse_file(), fsg_set_add(), fsg_set_select() and then ps_update_fsgset() I get very bad results (50%).
I think that pocket sphinx is kind of adapting the model at each recognition, maybe there is a way to reset that without loading everything again ?
Thank you for your help!
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hello,
I want to change my dictionary and my JSGF grammar file without reloading the acoustic model.
However it seems that I get very bad results if I do this without reinitializing the whole decoder. Indeed, if I reload everything I get an accuracy higher than 90% but if I only update the dictionary using ps_add_word(), nd change the JSGF using fsg_set_remove_byname(), jsgf_parse_file(), fsg_set_add(), fsg_set_select() and then ps_update_fsgset() I get very bad results (50%).
I think that pocket sphinx is kind of adapting the model at each recognition, maybe there is a way to reset that without loading everything again ?
Thank you for your help!
no
Yes
In order to let us help you with this problem you need to provide the data you are using for testing.
I finally solved it.
I made a mistake and switch the grammar too soon.